University of Redlands InSPIRe @ Redlands Undergraduate Honors Theses Theses, Dissertations & Honors Projects 2017 Analysis of Labor Market Inefficiencies in the NBA Draft Due to College Conference Bias Joshua C. Wall University of Redlands Follow this and additional works at: https://inspire.redlands.edu/cas_honors Recommended Citation Wall, J. C. (2017). Analysis of Labor Market Inefficiencies in the NBA Draft Due to College Conference Bias (Undergraduate honors thesis, University of Redlands). Retrieved from https://inspire.redlands.edu/cas_honors/170 This Open Access is brought to you for free and open access by the Theses, Dissertations & Honors Projects at InSPIRe @ Redlands. It has been accepted for inclusion in Undergraduate Honors Theses by an authorized administrator of InSPIRe @ Redlands. For more information, please contact rebecca_clayton@redlands.edu, paige_mann@redlands.edu.
Running head: NBA DRAFTING BIAS 1 Analysis of Labor Market Inefficiencies in the NBA Draft Due to College Conference Bias Joshua C. Wall University of Redlands Committee Chair: Nicholas Shunda, Department of Economics Committee Member: Jim Bentley, Department of Mathematics Committee Member: Nathaniel Cline, Department of Economics
NBA DRAFTING BIAS 2 Introduction "If you're going to lose, you have to lose because you can't be in the middle of the pack. You either have to be great or you have to be bad to get a good pick"(magic Johnson, 2014). As with any industry, success in the National Basketball Association (NBA) is deeply rooted in hiring the right person to do the right job at the right price. For example, a hiring manager would not pay an employee a substantial wage to be an engineer if the employee s only skills are as a dishwasher. However, NBA teams do not have as much freedom in choosing their players as most other typical private sector industries when hiring new employees. Nearly every major sport league in the United States holds an annual draft, which gives teams the opportunity to choose amongst the best of the amateur players and offer them contracts for future seasons. From an economics standpoint, these drafts are an effective way to observe the dynamics of a labor market in action. Since productivity in professional sports is relatively easier to measure then in other industries, professional sports give great opportunities to study the decision-making process of team owners and general managers. Every year all the major-league drafts are huge ordeals for the league, players and teams, but perhaps the most impactful draft on the league is the draft held by the NBA. The significant impact the draft has on the NBA compared to other major sports is because of how few players actually play at a time. In the National Football League, there are 11 players on the field at a time, with up to 53 on the team. Major League Baseball is similar with 9 players in the field at a time and up to 40 players on their expanded roster. In the NBA, there are only 5 players on the court for each team at a time, with a
NBA DRAFTING BIAS 3 maximum of 15 players on the roster. This means that a single great player could make the difference between a losing season and winning the NBA Finals. The purpose of my research is to determine what, if any, variables are significant in predicting a player s draft position as well as predicting their potential productivity in the NBA. I am specifically observing the effects of a player s college conferences effects on when a NBA will draft them. With the NBA draft being the biggest summer event for the league, it is not surprising that basketball fans and NBA teams alike take the day very seriously. The NBA draft dates back to 1947 (NBA.com), with nearly all Hall of Fame players having been brought into the league via the NBA draft, the draft is seen by most teams as an opportune time to hit the jackpot with a franchise changing player. However, despite for the draft being around for 70 years, teams still consistently draft players that just do not live up to the team s expectations. In today s league, this often leads to teams losing millions of dollars and missing out on an opportunity to draft the next NBA star. While this could be simply caused by bad luck, such as in 2007 when the Portland Trailblazers passed up NBA superstar Kevin Durant to draft Greg Oden who ended up getting an early career ending injury. But other times it is simply because the teams just did not see the potential in the player, such case as with Isaiah Thomas (drafted in 2012, not to be confused with the Hall of Famer from Detroit) who was drafted last in his draft class but has gone on to be voted into the league s All-Star game. The reverse order method of the draft, where teams with the worse winning records of the previous season are given the best chance of getting the top picks, is the league s attempt to distribute the talented players around the league and keeping all the teams as competitive with one another as possible. While there are many ways of
NBA DRAFTING BIAS 4 acquiring new players, whether it be through trade or signing free-agents, picking up a rookie via the NBA draft has often been considered, by most teams, one of the quickest ways of getting top players and turning a losing team into a championship team. Because of the importance put on the draft, at times fans will be eagerly hoping for their team to lose in hopes to get a lower draft position. Price, Soebbing, Berri, and Humphreys (2010) found suggestive evidence of teams purposely dropping games or shirking, specifically towards the end of the regular season when they have already been eliminated from the playoffs. But is the opportunity of a top draft pick truly worth it? In most cases, the number one thing that a team might look for in a player, as with employees in any industry, would be how productive they will be. In basketball, more productivity means more wins, more wins means a greater potential to go to the playoffs, which results in more games and more revenue via tickets and merchandise. The ultimate motivation for my research is to be able to isolate any key variables that NBA teams over or undervalue when taking into consideration which players they are going to draft. Knowing more specifically what to look for in players will not only help improve team productivity, but team profits as well. In order to answer my question, the assumption must be made that, as with any industry, you would expect the demand for the most productive players to be highest and therefore the most productive would be drafted first. If NBA teams do have any bias towards drafting players from certain college conferences or players with particular physical characteristics that do not have relevance in the player s ability to produce wins, it is likely that the NBA teams are unknowingly restricting themselves from putting together a winning team that would help boost the team s revenues.
NBA DRAFTING BIAS 5 Literature Review Starting in 1994 the NBA decided to restrict salaries of incoming rookies, which has since allowed teams to sign players, potentially worth tens of millions of dollars, for a fraction of the cost (NBA Collective Bargaining Agreement). Motomura et al. (2016) explore the effect of restricting rookie salaries and how the reduced cost of signing a young, unproven player compares to signing a proven free agent in the long run. The authors examined the improvement of teams who had more, higher, draft picks and relied more heavily on those picks over the course of four season. In general, Motomura et al. (2016) found results that suggested that teams reliant on their high draft picks were less effective than other teams who were able to find and develop players that matched their team s playing style. The ultimate conclusion proposed by the authors was that the teams that did not rely on draft picks were able to do so because their coaching staff was more capable of developing their own talent rather than relying on bringing on already talented players. The argument that coaches are more important in improving productivity than getting the top draft picks is supported by Arel and Thomas (2011), who studied the relationship the year a college student decides to enter the draft, and their potential success in the NBA. Their study found that higher draft picks do not generally help their teams win more games, rather the players picked later in the first round by better teams, which typically have better coaches, developed over time into players that are the most productive. However, there have been instances where single draft picks have come in and nearly overnight completely turned around an organization. For example, in the 2002-
NBA DRAFTING BIAS 6 2003 season the Cleveland Cavaliers won only 17 out of 82 games. This ultimately resulted with them getting the first draft pick in the 2003 draft, which they used to draft Lebron James. Immediately, the Cavaliers went from being a last place team, to consistently being a top competitor in the Eastern Conference. When James decided to take his talents to Miami in 2010, Cleveland dropped from the number 1 team in the east in 2009 to last place in a single season. Then upon Lebron s return to Cleveland in 2014, the Cavaliers went from not even making the playoffs, to back-to-back finals appearances with 2016 ending with them winning their first NBA championship in franchise history. (NBA.com) A team either wants to improve and win more games or gain a new superstar to be the face of their team. Drafting a well-known player from a major college, especially if they played well in the previous year s NCAA tournament, has been shown by an increase in the team s fan base. In fact, drafting a player that has Star Power (Berri, Schmidt, and Brook 2004), even if the team continues to lose, can have similar effects in terms of team revenues as if the team did start to win more games (Berri, Schmidt, and Brook 2004). The Star Power effect becomes even more profitable if the player performs well and can even start to become profitable for opposing teams. Hausman and Leonard (1997) estimated that at Michael Jordan s peak, through externality effects alone, would bring in nearly $50 million of revenues per season for other teams he was not playing for. A player s Star Power is typically generated by their ability to get major media coverage either by having an outstanding performance in the NCAA tournament, or more often attending a college that has regularly televised games. The Power 5, which includes
NBA DRAFTING BIAS 7 the ACC, SEC, Big-10, Big-12, and Pac-12 are currently the most popular conferences in the NCAA and get the most media converge by far (RPI Rankings). In 2015, the Power 5 conferences made $6 billion in total revenue, which was $4 billion more than all other NCAA schools combined, and most of the difference is revenues came from multimillion dollar media contracts and ticket sales (Lavigne 2016). Because of their mass media coverage, top players in the Power 5 often get Star Power from their conferences large fan bases, which as Hausman and Leonard (1997) would suggest, could be financially beneficial for NBA teams in terms of ticket and merchandise sales. Also, players that enter the NBA draft the same year they played in the NCAA tournament have been noted to have an improved draft position if they performed well as an individual or as a team (Ichniowski and Preston 2012). While the argument of wanting a high draft pick to draft a player with Star Power has some merit, Motomura et al. (2016) suggest that building a team through the draft is not the most effective way of building a winning team. Rather than focusing on a player s name recognition, focusing on learning a player s productivity potential throughout their career would give NBA teams a long term advantage over other teams. The question that then arises is that are teams currently focusing more on Star Power instead of player productivity. Since basketball is a complex, multi-person sport, which means evaluating an individual player s productivity based off one aspect of their game does not give a good overview of how much they actually contribute to the team. From the literature, the most commonly used method of measuring player productivity in basketball has been the Win Score Index, developed by Berri, Schmidt, and Brook (2006). The Win Score Index is
NBA DRAFTING BIAS 8 used in many different studies as the measurement of basketball productivity including by Coates and Oguntimein (2010) to predict career length and Ichniowski and Preston (2012) to relate NCAA tournament performance to draft position. The Win Score Index uses in-game statistics that either positively or negatively reflects on a player s performance. Statistics such as points, rebounds, steals, blocks, and assists were viewed as positive influences and so they will add to a player s Win Score. Turnovers, personal fouls, field goal attempts, and free throw attempts were viewed as negative influences and so they will subtract from a player s Win Score, which is defined as follows: Win Score Index = Points + Rebounds + Steals + ½*Blocks + ½*Assists Field Goal Attempts ½*Free Throw Attempts Turnovers ½*Personal Fouls. To build the simple Win Score Index, Berri, Schmidt, and Brook (2006) used data from the 1993-94, and 1994-95 NBA season to calculate the average points per possession in the NBA. Once finding the points per possession average was about 1, the authors determined the averages per possession for all the other major basketball statistics which included rebounds, steals, assists, blocks, field goal attempts, free throw attempts, turnovers and personal fouls. After calculating the per possession averages, Berri, Schmidt, and Brook (2006) then determined the ratio of each statistic over the average points per possession. The ratios have been reflected as the coefficients, which each statistic is pre-multiplied by to find the Win Score. For example, the per possession average for points was 1 and blocks was.5, so a player s game average of blocks will be multiplied by the ratio of.5/1 when calculating the Win Score. Win Score however has not been the only method of measuring basketball productivity in the NBA. Win Share is similar to Win Score in the sense that both
NBA DRAFTING BIAS 9 measurements are measured in terms of number of wins a player contributes to their team. Win Share takes the process a step further however and takes into consideration the statistics of the player s teammates and calculates each player s offensive and defensive scores independently. For example, when calculating a player s offensive Win Share, you must first find the individual player s points, offensive rebounds, assists, turnovers, field goals attempted and made, and free throws attempted and made. Then each statistic is weighted compared to the team s total for each individual statistic, and added together similar to Win Score by the statistic s ratio to points scored per possession. Defensive Win Share is then calculated in a similar manner but using steals, blocks, defensive stops, and defensive rebounds, and the two values are then added together to get a single Win Share value (Oliver 2011). Intuitively, the best way to predict a player s potential productivity and career success in the NBA would be to analyze their in-game statistics from their pre-draft performances because chances are a player that is not able to perform well at the college level will not be able to perform well in the NBA where the competition is much greater. Berri and Schmidt (2010) use players college in-game statistics to evaluate players WinScore in the NBA. The results reveal that in-game statistics such as rebounds, twopoint shooting percentage, and steals were positively associated with winscore, while points scored were negatively associated with winscore. Coates and Oguntimein (2010) obtain similar results but only for players coming from big college conferences, while points scored by players from smaller college conferences had a positive significant effect on productivity. Coates and Oguntimein classify big college conferences as those conferences they deemed as premier conferences and which included the Big-10,
NBA DRAFTING BIAS 10 Southwest, BigEast, Southeast, Metro, ACC, Pac-10, and Big-8. The conference effect is likely linked to the fact that players from the smaller conferences do not have the Star Power, and so players from the small conferences must be even more talented to be noticed by NBA teams. Model Specification The following equations represent the regression of winscore, winshare, and draft position(draftpos) models. The models will output estimated β coefficients that will give the estimated effect each independent variable will have on the dependent variables, all else equal. At the end of each model the ε term is the stochastic error term of the equations and β " at the beginning is the constant coefficient term. winscore + = β " + β. cpts + + β 1 creb + + β 3 cast + + β 5 ncaatw + + β 6 ncaachamps + + β 9 height + + β ; white + + β < age + + β = age + 1 + β." forward + + β.. center + + β.1 nbacoach + + β.3 ncaacoach + + β.5 movedcoach + + β.6 acc + + β.9 sec + + β.; big10 + + β.< big12 + + β.= pac10 + + β 1" bigeast + + ε + winshare + = β " + β. cpts + + β 1 creb + + β 3 cast + + β 5 ncaatw + + β 6 ncaachamps + + β 9 height + + β ; white + + β < age + + β = age + 1 + β." forward + + β.. center + + β.1 nbacoach + + β.3 ncaacoach + + β.5 movedcoach + + β.6 acc + + β.9 sec + + β.; big10 + + β.< big12 + + β.= pac10 + + β 1" bigeast + + ε + draftpos + = β " + β. cpts + + β 1 creb + + β 3 cast + + β 5 ncaatw + + β 6 ncaachamps + + β 9 height + + β ; white + + β < age + + β = age + 1 + β." forward + + β.. center + + β.1 ncaacoach + + β.3 movedcoach + + β.5 acc + + β.6 sec + + β.9 big10 + + β.; big12 + + β.< pac10 + + β.= bigeast + + ε + Each model includes anywhere from 19-20 independent variables, which have been sub-divided into four categories based on the nature of the variables that were being measured. The categories include college performance, physical characteristics, coaching, and college conference. College performance variables are intended to capture how well
NBA DRAFTING BIAS 11 the player performed in terms of their in-game statistics, as well as their post-season success, while in college. Physical characteristics include variables representing the player s physical attributes at the time they were drafted such as height, age, race, and the position they played on the court. The coaching category is designed to give any indication of how a player s coaches have effect their productivity and draft position; the category includes observing coaches from college as well as the NBA. The final category of college conference captures the effects of a player s college conference on their NBA productivity and draft position. College conferences were measured in a variety of ways including measuring specific conference membership with a series of dummy variables, grouping the major conferences into single dummy variable, splitting the major conference by their post-season success rate, and dividing the major conferences by their year of establishment. Winscore and winshare are both measuring player s NBA productivity in terms of average wins contributed to the team in the player s first four years in the NBA. The four year time-frame was selected because by the rookie contract rules, set by the NBA s Collective Bargaining Agreement, a rookie contract could extend to up to four years, depending on whether or not the team decides to offer the player the two seasonal extensions at the end of their second and third seasons. In the third model, draftpos represents the position that a player was selected in their respective draft. Important to note in the draftpos model is that negative coefficients are related to an improved draft position and positive coefficients are related to a worse draft position. The reversed signs of draft position are because of the values of the draft range from 1-60, with 1 being the
NBA DRAFTING BIAS 12 highest or best position a player can be chosen and 60 being the lowest or worse position a player can be drafted. The college performance category of variables includes the player s average points scored per game in college (cpts), average rebounds per game in college (creb), average assists per game in college (cast), total number of NCAA tournament games won the year before they were drafted (ncaatw), and a dummy variable representing whether or not their team won the NCAA championship (ncaachamps). The prior findings from Berri and Schmdit (2010) and Coates and Oguntimein (2010) would suggest that the signs of coefficients on ctps, creb, and cast should be expected to be positive in the winscore and winshare models and negative in the drafpos model. In addition, Berri and Schmidt s research found that ncaachamps has a negative effect on productivity, but a positive effect on draft position, so I should expect to see similar effects in my own results. Since there has not been any prior research on the effects of ncaatw on productivity, my prediction is that ncaatw will have positive effects on draftpos and no effect on winscore and winshare. The expectation of ncaatw having no effect on player productivity is because at most the player will play six extra games, which in the scope of how many the games and practices the player has already had, six additional games will not likely improve their basketball abilities. However, Ichniowski and Peston (2012) have found positive effects on player s draftpos, all else equal, likely caused by the player having more publicity. The additional publicity will likely cause a fan following and give the player some Star Power, which as suggested by Berri, Schmidt and Brook (2004)
NBA DRAFTING BIAS 13 Hausman and Leonard (1997) might incentivize a team to draft the player to gain their fans as well. The physical characteristics category of variables includes the player s age (age), height in inches (height), whether or not the player is white (white), a dummy variable indicating whether or not the player plays center (center), and a dummy variable indicating whether or not the player plays forward (forward). Since most of the physical characteristics could change over time, with the exception of the player s race, the physical characteristics are measured at the time the player was drafted. While it could be debated that physical characteristics such as height and age could have effects on a player s winscore and winshare because taller players have a physical advantage and young players would generally be expected to be less accident prone, none of the physical characteristics are likely to have any effect on productivity. Berri and Schmidt (2010) would support the claim of physical characteristics having no effect on productivity because in their results they concluded that the only physical characteristic to have significant results was playing center, which lead to less productivity in the NBA. However, Berri and Schmidt (2010) did find that height relative to position played and player s age at time of draft had positive and negative effects on player s draft position respectively. While Schroffel and Magee (2012) would argue that race would also have an effect on draftpos since their research revealed that players that are the same race as the head coach tend to get approximately 45-55 additional seconds of playing time per game, Berri and Schmidt s (2010) model is most similar to mine, so I would also expect for player race (white) to have no effect on winscore, winshare, or draftpos.
NBA DRAFTING BIAS 14 In addition to being in the model linearly, age will be included a second time as the non-linear variable age 2. Age 2 is included to capture any diminishing effects age might have on the winscore, winshare, or draftpos. While age is expected to have a negative coefficient in the draftpos model, age will likely have a much more extreme effect on draft position if the player was in the late 20s compared to their early 20s. For example, when a player is 21 years old, an additional year might expect to see the player to be drafted 2 positions later. But for a player who is 28, an additional year could possibly push them back 5 positions. The coaching variables that have been included in the model are whether or not the player played for a veteran NCAA coach (ncaacoach), whether or not the player played in a non-major conference but for a coach that used to coach in one of the major college conferences (movedcoach), and whether or not the player was drafted by a team that had a veteran NBA coach (nbacoach). In the case of NBA coaches, a veteran coach is defined as a coach with more than 10 trips to the playoffs with at least one NBA Finals appearance. A veteran NCAA coach is any coach with more than 10 trips to the NCAA tournament with two or more Final Four appearances (Tiernan 2013). However, because of the nature of the draftpos model, estimating draft positing by nbacoach would not make sense since the NBA coach has yet to have any impact on the player at the time of the draft. Because of the timing issue with nbacoach, nbacoach only appears in the winscore and winshare models. The signs of the coefficients for the coaching variables in the winscore and winshare models are expected to be positive. Motomura et al. (2016) concluded that coaches were more important to build a winning team than getting early draft picks. The
NBA DRAFTING BIAS 15 logic behind this conclusion is that better coaches are more knowledgeable and know what they are looking for in players. Since my definition of veteran coach implies success, both nbacoach and ncaacoach include coaches that have been successful and are likely better at picking productive players. The Movedcoach variable has competing hypotheses which means that the variable has differing effects depending on the signs of the variables estimated coefficients. These competing hypotheses are to help determine if college coaches are developing talented players, or are simply efficient at recruiting players that are already talented. The way I modeled movedcoach, if the beta coefficient is positive in the productivity models, it means that coaches seem to have a positive effect on player s productivity, all else equal, and do indeed help players develop their talent. Otherwise, if the movedcoach coefficient is positive, then it suggests that the coaches are not developing talent, but rather recruiting already talented players. I can the infer that if a coach is no longer coaching in the major conferences, but decided to continue to pursue a coaching career, the reason would likely be due to their in ability to make the major conference program successful either through their lack of ability to recruit or develop productive players. As a result, the coaches that are represented in movedcoach are forced to find other coaching opportunities at universities in lesser conferences and the movedcoach variable will likely have either a negative or zero effect on player productivity, all else equal. For the ncaacoach and movedcoach variables, the expected signs of the drafpos coefficients is a little more difficult to predict. Since the NBA teams likely do not have any information about the effectiveness of the coach a player had previously played for,
NBA DRAFTING BIAS 16 otherthan the coach s teams performance, NBA teams are unlikely to make any decision about how well they believe the coach coaches. Instead, NBA teams are likely to focus more on how well the team performs in the NCAA tournament and make their assumptions of the coach s ability from their tournament performance. This means that since ncaacoach is categorized by how well the coach has done in the NCAA tournament, the sign of ncaacoach coefficient will likely be associated with an improved draft position because NBA teams are likely to assume that the coach s consistent success in the NCAA tournament is due to their ability to develop their player s talent. Movedcoach will likely follow the same pattern as the ncaatw coefficient, once again because the only real way NBA teams are capable of measuring a coach s ability is by their team s performance. If movedcoach does have the same coefficient sign as ncaatw, I should expect to see that movedcoach will be associated with a better draft position, all else equal. The final independent variable category measures college conferences and is the main focus of the model, so the conference category is slightly more complex. The category is broken down into four versions, with each version observing the six major college conferences for basketball, which includes the ACC, SEC, Big-10, Big-12, Pac- 10 (now the Pac-12), and BigEast. The result will be four different versions of each model measuring winscore, winshare, and draftpos, for a total of twelve estimated models. The first version of the conference category keeps each college conference as a separate dummy variable, only measuring whether or not a player attended a college in that specific conference. For example, a player who attended Duke University, which is part of the ACC conference, would have a 1 denoted in the ACC variable and a 0 in the
NBA DRAFTING BIAS 17 other five conference variables. The purpose of keeping the conferences separate is to be able to compare specific conference significant effects in winscore, winshare, or draftpos. The second version of the conference category groups the six major conferences into a single dummy variable called major. Grouping the conferences into a single variable will give me the ability to observe whether or not as a whole the major conferences are preferred irrespective of which specific major conference a player comes from. The other two versions include splitting the six conferences by their post-season success rate and the year they were established. The Old vs. New version splits the conferences into two dummy variables, old conferences are those that were established prior to 1980 (ACC, SEC, Big-10, and Pac-10) and new conferences are those that were established after 1980 (Big-12 and BigEast). Top 3 vs. Next 3 splits the major conferences by their post season success, with the Top 3 representing the conferences with the highest win percentages in the NCAA tournament amongst the six major conferences (ACC, BigEast, and Big-10) and Next 3 representing the conferences with the lowest win percentages in the NCAA tournament amongst the six major conferences (SEC, Big-12, and Pac-10). The ACC, SEC, Big-10, Big-12, and Pac-10 were chosen because of their status of being part of the Power 5 and BigEast was included because of its exceptional post-season success in the last two decades, winning 6 of the last 20 NCAA championships. The breakdown of conference age and post-season success can be observed in Table 1 as follows.
NBA DRAFTING BIAS 18 Table 1: Major Conferences Conferences Established Overall Tournament Win Percentage ACC 1954 56.81% SEC 1933 47.71% Big10 1899 53.80% Big12 1996 47.53% Pac10 1916 50.34% BigEast 1980 56.52% The purpose of modeling conferences grouped whether or not they are a major conference by year of establishment and post-season success is to observe if specific characteristics of the conferences are what are appealing to NBA teams. While modeling each conference separetly can give an indication of NBA teams preferences of individual conferences, it does not give any indication of why the NBA teams have prefers that conference. The groupings add an additional layer to the model that could help suggest that NBA teams prefer a conference because they perform well, they are a major conference, or that the conference has simply been around for a long time. The expectation for the effects of college conferences is that any effect in the winscore and winshare model will be reflected by a similar effect in the draftpos model. If the NBA teams are rational in the selection process, meaning that they will draft the most productive players first, then if players from the ACC conference have statistically significant higher productivity, then players from the ACC conference should be shown to have statistically better draft positions. Assuming NBA teams are rational, the correlation between productivity and draft position by college conference would be consistent across all the different versions that the conferences are being measured, meaning that if an individual conference, or category of conferences, is associated with
NBA DRAFTING BIAS 19 higher levels of productivity, the conference should also have a similar associated effect in draft position. However, there could be some variation amongst the individual conference variables within each model. For example, in the separate conference models, there might be statistically significant positive effects from playing in the ACC, but no effects from playing in the SEC. An example of differing effects by college conferences comes from Coates and Oguntimein (2010), when their results showed differing effects depending whether or not the conference was big or small. Data The data used to estimate the regression models was entirely acquired from Basketball-Reference and strictly only observes players that played NCAA Division I basketball and were drafted between the years 2000 and 2010. All players that were drafted out of international leagues, community colleges, Division II universities, and colleges were excluded from the model in order to get a better indication of Division I conference effects on productivity and draft position. Of the 549 amateur players drafted in the 11 NBA draft between 2000-2010, (85%) were from Division I universities with only 97 players (18%) being drafted from non-division I universities. At the time of the drafts, the average player was a height of 6 6, weighed 222lbs, and was 22 years old. During their college careers they averaged 13.32 points, 5.71 rebounds, and 2.2 assists per game, while also averaging 1.5 wins in the NCAA tournament the season prior to entering the draft. In their first four years in the NBA the average winscore was 2.41, with a high of 12.50 set by Kevin Love, and a low of -1.7 by Troy Bell. The average winshare was 1.76, with a high of 13.38 set by Chris Paul, and a
NBA DRAFTING BIAS 20 low of -0.6 by Kenny Satterfield. The subtle differences between winscore and winshare are not too much of a concern because the two productivity measurements are strongly correlated. Meaning that even though winscore and winshare are measured differently both variables give similar results and we should expect relatively similar player rankings between the two measurements. Additional summary statistics about some of the key variables are given in Table 2, as well as their levels of correlation in Table 3. Table 2: Summary Statistics (1) (2) (3) (4) (5) VARIABLES N mean sd min max WinScore 2.413 2.310-1.700 12.50 WinShare 1.760 2.180-0.600 13.82 College Points/game 13.32 3.748 3.400 26.60 College Rebounds/game 5.716 2.079 1.400 12.70 College Assists/game 2.200 1.543 0.200 8.700 NCAA Tournament Wins 1.499 1.826 0 6 Height (inches) 78.90 3.413 68.90 87.01 Age (years) 21.94 1.462 19 30 Table 3: College Performance Correlation WinScore WinShare Draft Position College Points/game College Rebounds/game College Assists/game WinScore 1 WinShare 0.886 1 Draft Position -0.5515-0.5097 1 College Points/game College Rebounds/game College Assists/game 0.1549 0.2093-0.2677 1 0.3879 0.2201-0.172 0.2466 1-0.0131 0.1065-0.0778 0.237-0.4448 1
NBA DRAFTING BIAS 21 With conferences being one of the main focal points of my model, being able to see any immediate difference amongst conferences could give an early indication of what to expect in the inferential results. Table 4, which compares the representation of drafted players from each conference and each conferences total representation of the NCAA, indicates an over-representation of players being drafted from the major conferences. Each conference represents less than 5% of all the NCAA Division I teams but also each conference is represented by over 10% of the drafted players, with the exception of the Big-10. In total, over 75% of the Division I players drafted were from one of the major college conferences despite the conferences only representing 20.8% of all the Division I teams. The over-representation of major conference players in the draft suggests that NBA teams might have a preference to drafting major conference players. The preference would likely be caused by either the most productive players attend the college at the major conferences or that the major conferences are given the most publicity so that their players have more exposure to NBA teams and scouts. Table 4: Conference Representation Conference Percentage of Percentage of Drafted Players NCAA Players ACC 14.9% 3.42% SEC 12.1% 3.42% Big-10 8.9% 3.13% Big-12 11.9% 3.42% Pac-10 14.6% 2.85% BigEast 14.2% 4.56% Major 76.6% 20.80% Other 23.4% 79.20%
NBA DRAFTING BIAS 22 Table 5: Mean Statistics by Conference (1) (2) (3) (4) (5) (6) VARIABLES ACC SEC Big10 Big12 Pac10 BigEast WinScore 2.720 2.880 2.050 2.390 2.820 2.310 WinShare 2.040 1.920 1.550 1.710 2.360 1.510 Draft Position 26.76 30.77 27.14 27.21 28.25 27.85 College Points/game 13.13 12.27 13.30 13.30 13.06 12.56 College Rebounds/game 5. 5.660 5.279 5.936 5.697 6.040 Table 5 give summary statistics broken down by each individual major conference. Across the individual conferences 14.9% of the drafted players were from the ACC, 12.1% from the SEC, 8.9% from the Big-10, 11.9% from the Big-12, 14.6% from the Pac-10, and 14.2% from the BigEast. In terms of college performance, the average points and rebounds was consistent in every conference with average points being around 13 points per game and average rebounds being 5 rebounds per game. Productivity was not as consistent across conferences, the average winscore by each conference was broken down into two groupings with the ACC, SEC, and Pac-10 having an average winscore around 2.7 and the Big-10, Big-12, and BigEast averaging slightly lower at 2.3. Averages winshare scores followed a similar pattern as winscore with ACC, SEC, and Pac-10 averaging around 2 and the Big-10, Big-12, and BigEast being slightly lower at 1.6. Draftpos seemed to be the only key variable that was not evenly split, while the average draft position of players from major conferences was 27, the Pac-10 was an outlier with an average draft position of 30. The fact that of Pac-10 players average lower draft positions is significant because the same Pac-10 players also averaged slightly higher productivity levels. While the information at this point is not enough to make any
NBA DRAFTING BIAS 23 conclusion, the data summary does give further indication of possible conference bias in the NBA draft. Table 6: Mean Statistics by Major Conference (2) (3) VARIABLES Major Other WinScore 2.560 1.930 WinShare 1.880 1.380 Draft Position 27.99 32.04 College Points/game 12.92 14.64 College Rebounds/game 5.699 5.771 Table 7: Mean Statistics by Old vs. New Conferences (1) (2) VARIABLES Old New WinScore 2.670 2.350 WinShare 2.020 1.600 Draft Position 28.22 27.56 College Points/game 12.93 12.90 College Rebounds/game 5.548 5.993 Table 8: Mean Statistics by Conference Performance (1) (2) VARIABLES Top 3 Next 3 WinScore 2.410 2.700 WinShare 1.730 2.020 Draft Position 27.26 28.72 College Points/game 12.96 12.89 College Rebounds/game 5.639 5.759 Tables 6-8 include further summary stats, but rather than being broken down simply by individual conferences, the tables are separated by the different groupings of conferences used in the regression models. Table 6, which includes players that played in
NBA DRAFTING BIAS 24 any major conference, gives good overall observations of the major conferences, revealing that the averages for the individual conferences were good indications of all the major conferences. Table 7 summarizes the conferences by conferences age and Table 8 summarizes the conferences by NCAA tournament performance. In general, players from older conferences seemed to have higher productivity and college performance averages; the exception is draft position, which players from the newer conferences average slightly lower draft position. A similar story can be seen comparing the conferences by performance, the top conferences tend to have lower productivity levels while also having lower draft positions, on average. Results Tables 9, 11, and 13, in the appendix, report the estimated regressions modeling player s college performance, physical characteristics, college conference, and coaches to explain winscore, winshare, and draftpos. The models use OLS to estimate the variables coefficients, and corrects for any heteroskedasticity by using robust standard errors. The decision to use robust standard errors was to correct for the heteroskedasticity in the winscore model that was found after using the white test on the original models. Tables 10, 12, and 14 report results of the F-Test for each variable category and their joint significance with one another. Each model has four versions that vary according to how college conferences are measured. In each table the first model represents the individual conference model, followed by the major conferences being grouped into a single variable, then split by conferences established before and after 1980, and finally divided by the conferences post season success rate.
NBA DRAFTING BIAS 25 Since the dataset does cover over a decade of statistics, there was concern about some type of serial correlation causing an error in the model. This would likely be caused if some of the draft classes were significantly better than others or if there were any rule changes that drastically changed the game. Specifically, during the 2006 season the NBA added age limit to be eligible for the draft, which eliminated any high school players from entering the draft. Without the addition of the top high school players, it is likely that college players were drafted that ordinarily would not have been, possibly causing a significant change in the effects on productivity and draft position. To test for any discrepancies caused by the year, I ran the regressions including dummy variables indicating the year that each player was drafted. Tables 15, 16, 17 in the appendix show the estimated results when the year effects were included. Comparing the coefficients of the models with and without years revealed little difference, hence suggesting that the model does not suffer from serial correlation. As shown in Table 9, in the winscore models, creb, cast, and ncaatw all are associated with a statistically significant positive effect on a player s winscore in each model. Age, nbacoach, and ncaacoach all are associated with a statistically significant negative effect on a player s winscore in each model. In the separate conference model the only conference to have any associated significant effect was the SEC, which playing in the SEC was associated with an improved player winscore by 0.703, all else equal. In the Old vs. New conference model, dividing the conferences by their established year made center as well as the old conference variables become statistically significant. In Old vs. New conference model, player s that played center had an increase of 0.748 in
NBA DRAFTING BIAS 26 their winscore, all else equal, and an increase of 0.390 if they played in one of the older conferences, all else equal. WinScore Table 10: WinScore F-Test Results Separate Major Old vs. New Top 3 vs. Next 3 90% 95% 99% 90% 95% 99% 90% 95% 99% 90% 95% 99% College Performance X X X X Physical Characteristics X X X X Conference X X Coach X X X X Table 10 reports the joint significance of each sub-category of variables with respect to the winscore models. For all four versions of the winscore model college performance and physical characteristics were significant at the 1% level. The conference and coaching categories varied between the model versions, with conference having no significance in the major conference model and Top 3 vs. Next 3 model, while being significant at the 5% level in the Old vs. New conference and significant at the 10% level in the separate conference model. The coaching category was significant at the 10% level for separate and Old vs. New conferences and 5% level for the major conference and Top 3 vs. Next 3 conferences. As shown in Table 11 of the appendix, the winshare models had college rebounds, college assists, and NCAA tournament wins as positively statistically significant variables and age as a negatively statistically significant variable to determine player s winshare. The winshare models also reported the addition of college points having a statistically significant positive effect in all versions of the models, every additional point averaged in college is associated with an increase in a player s NBA winshare by
NBA DRAFTING BIAS 27 approximately 0.077, all else equal. Observing the separate conference version winshare model, it can be noted the rather than the SEC being statistically significant like in the winscore model the Pac-10 has a statistically significant positive effect. The Pac-10 s effect on player winshare is slightly lower than SEC s effect on winscore, players who attended a college in the Pac-10 could expect to have an associated increase of 0.599 in their winshare, all else equal. The other significant variable that can be observed is in the Old vs. New Conference model and is the variable representing if the player played in an older conference. Similar to the old variable s effect on winscore, a player from an older conference is likely to have a 0.392 increase in their winshare, all else equal. WinShare Table 12: WinShare F-Test Results Separate Major Old vs. New Top 3 vs. Next 3 90% 95% 99% 90% 95% 99% 90% 95% 99% 90% 95% 99% College Performance X X X X Physical Characteristics X X X X Conference X X Coach Table 12 reports the joint significance of each sub-category of variables with respect to the winshare models. For the winshare model the only consistent categories were college performance and coaching, while college performance was significant at the 1% level in all four versions coaching was not significant in any of the winshare models. Physical characteristics was significant at the 5% level in the separate conference and Old vs. New models, had significant at the 10% level in the major conference and Top 3 vs. Next 3 conference models. For the conference category, conferences were jointly significant at the 10% level and 5% level in the separate and Old vs. New conference
NBA DRAFTING BIAS 28 models respectively, and had no significance in the major and Top 3 vs. Next 3 conference models. The coaching category had no joint significance in any of the four versions of the winshare model. Before reviewing the draft position models in Table 13 of the appendix, it is important to recall the inverse nature of the Draft Position model. Since the values of draft position range from 1-60 with 1 being the best position and 60 being the worst position, negative coefficients represent an improvement in a player s draft position. For example, in all four versions of the model, college points have a negative statistically significant effect. College points effect on draft position is reflected in table as a negative coefficient of approximately -1.45. Because of the nature of the draft position model a - 1.45 coefficient means that for every additional point averaged in college, a player is expected to have an effect associated with an improved draft position by 1.45 position, all else equal. In other words, a player that would have been drafted 15 th will now be drafted 14 th or 13 th if they average one additional point in college, holding all other variables constant. In addition to college points, other variables that associated with improving player s draft position across all the versions of the model included college assists, NCAA tournament wins, and player height in inches. The only variable associated with a worse draft positon across all versions of the draft position model was age, which is associated with a decrease in draft position across all versions by approximately 11 positions for every year older the player was when they entered the draft, all else equal. In the separate conference version, if the player won the NCAA championship the year they were drafted, the player was expected to see a worse draft position by approximately
NBA DRAFTING BIAS 29 5.8 positions, all else equal. In addition, in the separate conference model, players from the ACC were expected to be drafted 4.3 positions earlier, all else equal, and the ACC was the only conference variable in the separate conference version of the draft position model associated with an improved draft position. The other conference variables with association with better draft positions in the draft position model included players from newer conferences in the Old vs. New version, and players from one of the Top 3 conferences in the Top 3 vs. Next 3 Conferences model. In both cases of new and Top 3, a player from a conference fitting in one of these categories was expected to improve their draft position by approximately 3.4 positions, all else equal. Draftpos Table 14: Draft Position F-Test Results Separate Major Old vs. New Top 3 vs. Next 3 90% 95% 99% 90% 95% 99% 90% 95% 99% 90% 95% 99% College Performance X X X X Physical Characteristics Conference Coach X X X X Table 14 reports the joint significance of each sub-category of variables with respect to the draft position models. In the draft position models, all four categories had consistent joint significant levels. College performance and physical characteristics were both significant at the 1% level, and college conference and coaching were not significant at any reasonable level in all four versions of the draft position model.