The length and success of NBA careers: does college production predict professional outcomes?

Coates, Dennis ; Oguntimein, Babatunde

Introduction

Can professional sports franchises identify quality players based on the information available to them prior to a draft? Do the attributes that determine draft position also indicate career success? Do those attributes imply something about what the drafting teams value and are they the same attributes that professional teams compensate highly? At the heart of these questions is an important issue in all of economics: Are economic decision makers rational? The answer to this question is, of course, key to understanding the efficiency properties of the player labor market. In this paper we use various college performance statistics to address each of these questions. Our focus is on basketball because the statistical measures of individual production are cleaner in basketball than in football, and are far more available and comparable across players at the pre-draft level for basketball than for baseball.

A variety of papers have addressed the effectiveness of teams to select quality players during the annual amateur drafts held by professional sports leagues. The Major League Baseball draft is studied by Spurr (2000), the National Football League draft by Hendricks, DeBrock, and Koenker (2003), Massey and Thaler (2006), and Boulier, Stekler, Coburn, and Rankins (2004). Studies of the NBA draft are Staw and Hoang (1995), Camerer and Weber (1999), Groothuis, Hill, and Perri (2007a, 2007b), and Berri, Brook, and Fenn (2008).

Hendricks, DeBrock, and Koenker (2003) use the NFL draft to assess the way in which signals about uncertain productivity influence hiring decisions. They argue that uncertainty influences the decision in two ways, via statistical discrimination, in which a highly inefficient signal about future productivity puts one group at a disadvantage, or via option value, in which firms give members of the riskier group a probationary period in which to demonstrate their productivity. They find support for both statistical discrimination and option value effects in the NFL draft. Players taken early in the draft from less-visible college football programs tend to have better careers than their counterparts from the highly regarded programs. Comparing two players, early in the draft the NFL teams favor the one from the more highly ranked college program. Later in the draft, players from the higher ranked programs are undervalued, suggesting an option value explanation. The authors note that the NFL has an enormous amount of information about the players, which should make systematic errors in hiring less likely. Their evidence indicates that despite all this information, systematic errors are common.

Massey and Thaler (2006) also examine the NFL draft. The question they ask is whether NFL franchises accurately value draft position. They begin be assessing the value of one draft pick relative to another by using draft day trades of higher draft picks for lower ones. They find that the relative value of draft picks drops off very quickly and that future draft picks are discounted at extremely high rates. Following this, they evaluate the performance of players drafted at different positions using a variety of measures of performance. They find that performance does not drop nearly so steeply with draft position as does the value of draft position relative to the top draft position. They find that high picks are overvalued, costing both a large number of lower draft choices and a high salary, neither of which corresponds well with the added performance from those positions.

In their study of the NBA draft, Staw and Hoang (1995) assess the draft position of a player and that position's relation to playing time and survival in the league. Their analysis is a test of whether and how people respond to sunk costs, in this case the use of a high draft pick and the money to sign that player to a contract. Their evidence is that franchises are more likely to stick with high draft choices longer than low draft choices, even after controlling for productivity, an activity described as escalation of commitment. One of their questions is specifically whether draft position determines career length. Their data set covers a subset of players from the 1980 through 1986 drafts and follows them through the 1990-91 season. The players must have been drafted in one of the first two rounds and stayed in the league for one year or more. When their data end in 1991, 91 players drafted in the first two rounds between 1980 and 1986 are still playing in the NBA.

Camerer and Weber (1999) also address the issue of escalation of commitment. Their concern is that Staw and Hoang did not rule out alternative explanations for the apparent escalation of commitment to early draft choices. For example, Camerer and Weber suggest that the productivity of alternative players may be worse still than the drafted player, in which case it would be rational to stick with the draftee. Using a different sample, from more recent seasons, but following the Staw and Hoang methodology, Camerer and Weber replicate the Staw and Hoang results. They then address other explanations for the persistence of draft position on playing time and productivity. They conclude that their tests reduce the magnitude of the escalation but that escalation persists.

Spurr (2000) examined the MLB draft for the ability of baseball clubs to identify talent. The MLB draft is different from the NFL and NBA drafts in three important respects. First, the MLB draft includes players right out of high school and, in fact, has only evolved toward predominance of college players in the last 20 years, whereas draft of high school players is not allowed in the NFL or the NBA. Second, the MLB draft continues for far more rounds than either the NFL or NBA drafts. Currently, the MLB draft stops at 50 rounds, but it had been unlimited until 1997. The NFL draft currently lasts seven rounds, down from 12 as recently as 1992, and the NBA draft is now limited to two rounds, down from 7 in 1987. The third major difference between the MLB draft and those for the NFL and NBA is that baseball draftees do not go immediately to the big leagues as the football and basketball players do. Instead, players drafted by baseball teams go into an extensive minor league system where they develop their skills before rising to a major league roster. Winfree and Molitor (2007) report that between 1965 and 1980 the average time in the minor leagues was 2.7 years.

Spurr's (2000) results showed that the probability of making the major leagues declines with draft position. His analysis also focuses on player background, whether drafted out of high school, community college, or a four-year college, player position, and drafting club. His results do not indicate any general effects of drafting club on the likelihood of making the majors, but he does find that college players have a higher probability of making the majors than do those drafted out of high school. Groothuis and Hill (2004) estimate a hazard model of the career of NBA players. They find that the probability of a career ending with the just completed season is higher the later in the draft a player was taken and that more productive players are more likely to be on a team the next season. They do not address the issue of whether a drafted player makes an NBA roster.

Groothuis, Hill, and Perri (2007b) study NBA performance for players between the 1987-88 and 2003-2004 seasons to evaluate the ability of teams to identify superstar players. They identify players whose NBA efficiency is more than three standard deviations above the mean efficiency for a given year. Few players ever achieve this and fewer still achieve it multiple times. In any given year, between 12 and 22 players have an efficiency greater than two standard deviations above the mean. Groothuis, Hill, and Perri estimate a random effects model in which a player's efficiency is a function of his draft position, experience and experience squared, height and weight, years of college, and race. They find that draft position is negatively related to NBA efficiency in a given year. Among their explanatory variables, only height and weight are not statistically significant.

Berri, Brook, and Fenn (2008) is closest in spirit to the analysis here. Focusing on more recent draft classes for the NBA, they assess the ability of collegiate performance to predict draft position. They find scoring and shooting efficiency to be significant determinants of where a player is drafted.

Academic literature has also addressed the efficiency of the players labor market by relating compensation to performance. We do not review this vast literature but focus on those studies related to basketball and to our basic question. Staw and Hoang (1995) and Berri, Brook, and Schmidt (2007) found that points scored is a--or the--dominant factor in assessing NBA player productivity. Staw and Hoang reported that a one standard deviation increase in their scoring index (a combination of points per minute, field goal percentage, and free throw percentage) resulted in a 4.6 year increase in career length. The scoring variable is statistically significant with a p-value less than 0.001. By contrast, a one standard deviation increase in the "toughness" index (rebounds and blocked shots) has an effect less than half that of scoring and the variable had a p-value of less than 0.05. Assists and steals (the quickness index) was not statistically significant.

Berri, Brook, and Schmidt (2007) conclude, "Player evaluation in the NBA seems overly focused upon scoring. Negative actions, such as inaccurate shooting or accumulating turnovers, do not seem to result in corresponding declines in player compensation." Berri, Brook, and Schmidt (2006) say that points are so important to NBA compensation, and shooting accuracy unimportant, that a player interested in maximizing his salary should "focus solely on chucking up as many shots as a coach allows" (p. 209). Interestingly, Kahn and Sherer (1988) found, for a sample of 226 players drafted before the 1985-86 season, that draft position was significantly affected only by number of college seasons played, college games per season, number of times selected to The Sporting News first or second team or winning College Player of the Year, and whether the athlete left college early. Neither points nor shooting percentage or any other specific performance measure was found to affect draft position. Consequently, we ask whether draft position is or is not adversely affected by poor shooting and turnovers in college and whether or not it is improved by increasing points, rebounds, blocked shots, and assists.

We also ask what relationship there is between college productivity and NBA productivity. Groothuis, Hill, and Perri (2007) related NBA efficiency to draft position and Berri, Brook, and Schmidt (2007) relate efficiency to pay, but neither of the studies examined college performance as a determinant of draft position or as an indicator of NBA success. We use college performance to predict draft position and NBA success. If NBA talent evaluators are successful, then those college statistics which influence draft position should be correlated with those same statistics in the NBA.

The analysis first evaluates draft position. We use career college statistics, points, rebounds, blocks, assists, steals, and personal fouls, per college minute played, the college career field goal, free throw and turnover percentages, and the NBA formula for computing productivity, as well as the Berri, Schmidt, and Brook (2006) Win Score measure of productivity, for college, to predict draft position. We also interact performance with an indicator for a top college basketball conference and with one indicating players from institutions whose teams played for a national championship in the 11 years prior to their draft year to allow for the possible statistical discrimination and option value issues raised by Hendricks, DeBrock, and Koenker (2003). The evidence suggests that scoring, rebounding, and blocking shots each individually improves draft position for players from what we term big conferences while committing fouls worsens their draft position. Players from lesser conferences or schools not aligned with conferences have their draft position improved by good shooting from the floor and from the free throw line, but not by any of the other performance statistics. Interestingly, pooling the small and big conference players is rejected by the data, indicating that NBA clubs evaluate performance of players from big and small conferences differently.

Next, we address NBA productivity. We estimate a probit model to predict whether a player makes an NBA team, and a tobit and negative binomial model relating draft position to length of NBA career. Given a player made an NBA roster, we examine the relationship between various measures of college productivity and the analogous measure of NBA productivity. For example, we use least squares regressions to relate college points (rebounds, assists, etc.) per minute to NBA points (rebounds, assists, etc.) per minute. Likewise, we relate college field goal and free throw percentages to NBA percentages. Finally, we relate college values for efficiency to NBA career values for efficiency. Our results suggest that some college statistics do well at predicting NBA statistics, and others do not. Moreover, there is some evidence that players from big conferences or highly regarded college programs are more productive over their NBA careers than players from smaller conferences.

The rest of this paper is divided into three sections. In the next part, we describe our data. Following that, we present our results. The paper ends with a summary of the results and a discussion of the implications for the ability of teams to predict which college players will be successful and the importance of noisy signals about player ability.

Data Description

Our data set consists of players from the draft classes between the years 1987-1989. We chose this span to study players whose professional careers had already been completed yet would be quite recent. Studying retired players makes it possible to consider whether college production can be used to predict successful NBA careers. For each player drafted between 1987 and 1989 who played at least one year in the NBA we collected draft year, college, conference, big conference, class, height, weight, position, points, rebounds, assists, steals, blocks, free throws attempted and made, field goals attempted and made, turnovers, games, minutes played, personal fouls, and three different efficiency statistics. We gathered the same variables for their NBA careers and also their draft round, draft position, and drafting team and years in the league. College statistics for players who made the NBA, and their NBA statistics, are available at http://www.basketball-reference.com and http://www.databasebasketball.com.

Our data overlaps that of Camerer and Weber (1999) in that we use the 1987 through 1989 drafts while they use the 1986 through 1991 drafts. They limit their attention to the first two rounds, but we include all draftees. In addition, their data, and that of Staw and Hoang (1995), is annual while we focus on career statistics. Staw and Hoang and Camerer and Weber both find that draft position effects decline over time, with each position in the draft reducing minutes played in a season by 22 minutes in the second season and by 11 minutes in the fifth season. Neither extends the analysis beyond the fifth season. We look at effects over the entire career.

We created a variable called big conference to identify players whose college was a member of a premier basketball conference. This is more specific than Hendricks, DeBrock, and Koenker's (2003) Division I-A variable but less specific than their top 30 variable. We define big conference equal to 1 for a college in any of the Big 10, Southwest, Big East, Southeast, Metro, Atlantic Coast, Pac-10, and Big 8 conference; big conference equals zero otherwise.

Given the college and NBA career statistics we computed two efficiency scores for the players' college and professional careers. (1) These are the efficiency formula used by the NBA and the Win Score measure proposed by Berri, Schmidt, and Brook (2006).

Table 1 reports descriptive statistics for the full sample and split into big conference and small conference players. Asterisks by variables identify those variables whose means are different between big and small conference players. We must emphasize that we only have college statistics for drafted players who played in the NBA, so these statistics may be upward biased relative to the means of all drafted players. The results show that players from small conferences had statistically significantly larger collegiate efficiencies per minute played than players from the big conferences. Players from small conferences were, on average, drafted with later picks than players from big conferences. These results are consistent with the statistical discrimination and option value arguments of Hendricks, DeBrock, and Koenker (2003). Moreover, comparing the mean values of efficiencies or points, rebounds, etc., between late round small and big conference players, one finds that the average for the small conference players is even bigger relative to the average for the big conference players than appears in the table.

Players drafted from small conferences who made it to the NBA have significantly shorter average careers than players from the bigger conferences. The smaller conference players that make it to the NBA do not have significantly different NBA production than the big conference players in the NBA. This finding is consistent with the Staw and Hoang (1995) and Camerer and Weber (1999) findings that franchises stick with early draft picks longer than late draft picks even after controlling for productivity.

Results

Table 2 reports regressions using college productivity to explain draft position. (2) For the statistically significant variables, the elasticity is reported below the standard error. Table A1 in the appendix reports results for the same regressions adding the player's height in inches and dummy variables for position played. Since none of these is individually nor are they jointly significant, and their inclusion has no meaningful impact on the college productivity variables, we do not discuss these results further. The sample is split between players from big conferences and those from conferences not deemed as big or from colleges not in a conference. The determinants of draft position for big conference players appear to differ from those for small conference players. (3) For example, field goal and free throw shooting percentages are important determinants of draft position for the small conference players while neither is a significant determinant of draft position for big conference players. For both of these variables, a one percent improvement in shooting accuracy produces about a 3.3 percent improvement in draft position.

In the big conference data, rebounds, points, blocks and personal fouls per minute are individually significant determinants of draft position. Among these individually significant coefficients in the big conference equation, the coefficient on points per minute has the expected negative sign while the same variable in the small conference equation does not. The other variables have the same sign in both equations, but only the blocks per minute variable is similar in size for small and big conference players alike. Testing for pooling of the two subsamples easily rejects the null hypothesis that all coefficients are equal across the two groups; the F-statistic is 3.903 with a p-value of 0.000.

The largest elasticity of draft position with respect to college productivity among the big conference players is that for personal fouls per minute. An additional one percent more fouls per minute lowers draft position by 1.58 percent. Points per game has the second largest elasticity, with one percent more points per minute improving draft position of a big conference player by 1.30 percent. Both of these are small in comparison to the elasticity of draft position with respect to shooting accuracy for the small conference players. This evidence indicates that college productivity affects draft position differently for players from big and small conferences. These results suggest that NBA franchises try to account for quality of competition when assessing draftees.

One might contend that big versus small conference players is not a fine enough distinction in the quality of the signal about player ability. Hendricks, DeBrock, and Koenker (2003) used Division I-A football players, a broader signal than our big conference, and the number of years between 1980 and 1992 that a draftee's college ended the season ranked in the top 30 to identify players whose ability signal was most clear. We tried a second measure of signal quality, namely whether the player's team was either national champion or played in the championship game during the 11 years prior to the draft in which the player was taken. This is surely too narrow a signal about the player's ability. In any case, one rejects pooling of the players from schools that participated in a national championship game with players from schools that did not. In this data, no small conference school played in the national championship game in the 11 years prior. Interestingly, the results from this analysis, reported in Table 3, are very similar to those distinguishing the big and small conference players. For players from schools that did not play in the national championship game, there is a premium on good shooting as both the field goal and free throw percentages are significant at the 1% level. By contrast, neither shooting statistic of players from national championship participating schools is significant at even the 5% level, though free throw shooting is significant at the 10% level. The draft position of players from championship game participant schools is influenced by scoring and rebounding while that is not true for players from non-championship game participant institutions. Blocking shots has about the same impact in both subsamples though it is only significant for players from non-championship participating schools, and that only at the 10% level. Committing fouls, on the other hand, is only significant in the championship participant sub-group. Turnover percent is significant but of the wrong sign for players whose institutions participated in the NCAA championship game. This puzzling result may be an artifact of the small number of observations in the regression. Testing for pooling between the championship and non-championship participant players, the implication is that draft position of these two groups are influenced differently by collegiate performance.

The next question is the extent to which draft position determines professional career success. Table 4 reports probit, tobit, and negative binomial regressions on all draftees in the sample. Because college statistics are only available for players who played in the NBA, using all draftees, as in Table 4, limits the independent variables available for the analysis. Table 5 shows results from tobit and negative binomial regressions explaining career length expanding the available explanatory variables but limiting the analysis to those draftees from our sample who played in the NBA and in college.

To have a successful career, one must first have a career, so we begin by estimating a probit model of the likelihood a drafted player ever plays in the NBA, reported in the first column of Table 4. We use draft position, draft position squared, big conference, and big conference interacted with draft position as our regressors. This approach is consistent with Camerer and Weber's (1999) argument that draft position contains information about expected productivity or other player attributes that are not observed by actual productivity measures and/or is a good index of the available information about a player. Significance of the big conference and interaction variables would be support for the statistical discrimination and option value arguments of Hendricks, DeBrock, and Koenker (2003).

The results indicate that draft position influences the likelihood of playing in the NBA, with being taken later reducing the probability of playing at least one season. Significance of the squared draft position term, with the positive sign, reveals that the marginal effect of draft position gets smaller in absolute value as draft position gets larger. Groothuis and Hill (2004) find that the probability of a player's career continuing one more season falls the later in the draft they are taken. They do not include a second order draft position term. Neither the big conference nor the big conference-draft position interaction is statistically significant. They are not jointly significant either. (3) These results are not consistent with the findings of Hendricks, DeBrock, and Koenker for the NFL draftees that suggest teams exercise an option by drafting players whose future performance is less certain but who may have great potential. (4) Tobit results in the second column, with many observations at zero years, also have draft position and draft position squared as the only significant variables.

The negative binomial model specification, ideal for count data, is reported in the third column of Table 4. The results are quite different than those of the probit and tobit. Specifically, while the estimates in the negative binomial model indicate that being drafted early increases the players expected career length, as in the former regressions, the marginal effect of draft position is not diminishing as draft position rises (draft position squared is not different from zero). Moreover, a player drafted from a "big conference" school has an impact from draft position about 50% larger than that for a small conference draftee; the coefficient on draft position for a small conference player is -0.024, which changes to -0.037 (-0.024-0.013) for a big conference player. In addition, independent of the extra boost to career length a big conference player can expect from a given position in the draft, big conference players get a career longevity bonus over players from small conferences as evidenced by the coefficient on the big conference dummy of 0.461 (significant at the 10% level).

Now we turn to the question of whether NBA career longevity can be predicted from college productivity, conditional on playing during at least one season in the NBA. A tobit regression is reported in the first column of Table 5, a negative binomial regression in the second column. As in the full set of draftees, evidence from the sample of players with NBA experience indicates that draft position is statistically significant and that players drafted later have shorter NBA careers. In both the negative binomial and the tobit models, there is a diminishing marginal impact of draft position just like in the probit and tobit models of Table 4. Neither big conference nor the big conference-draft position interaction terms are significant in either the Tobit or the negative binomial in Table 5. Among the college productivity statistics, only college free throw percentage is statistically significant in determining career length among those players who make it to the NBA. Oddly, a higher free throw percentage lowers the expected career length. This data also indicates that guards have a slightly longer expected career than forward or centers. The variable is only significant at the 10% level.

The results in Table 5 indicate that little can be gleaned about the likely duration of a player's NBA career from his collegiate productivity. (4) It may be that draft position and, to a lesser extent, the player's college conference capture all the relevant information about a player's likely longevity as a professional. But it may be that despite performance statistics not predicting career length, perhaps they will predict career performance. We turn to this question now.

One measure of performance is playing time. Players considered to be making valuable contributions to winning will play more than players making lesser contributions. Our analysis is similar to the Staw and Hoang (1995) and Camerer and Weber (1999) analysis, where they explain minutes played during a season. We explain minutes per game over the course of the entire career. Our results, explaining minutes per game during the entire career with draft position, draft position squared, and position dummies, are in Table 6. The results show that draft position is a significant determinant of minutes played over the course of a player's career for those players whose career is five or fewer years long. For players whose career is longer than five years, draft position does not predict playing time. While draft position is significant in the full sample, a Chow test rejects pooling the less than or equal to five year career and greater than five year career players. The coefficients for the less than or equal to five year career players indicate that a one percent drop in draft position translates into about a .20 percent reduction in minutes per game.

Camerer and Weber (1999) added other measures of productivity to the minutes played equation. In unreported regressions we add college points, rebounds, assists, blocks, and steals per minute, field goal, free throw and turnover percentages, with no effect, or College Win Score or the College value of NBA productivity, which are also not significant. College performance is unable to predict NBA playing time in either career length sub-sample or in the full sample.

Tables 7.1 through 7.4 report the results of regressions, which relate some collegiate performance statistic, say points per minute, to the points per minute recorded by the player as a professional. The regressions also include draft number and are estimated on the full sample as well as on the big conference and small conference player samples, respectively. The upper panel of Table 7.1 shows the results of the points per minute estimation. The results indicate that earlier draftees will score more points per minute as pros than will players drafted later, though the effect is not statistically significant for players from big conferences. College points per minute is positive and statistically significant in the full sample and for big conference players, but not statistically significant for small conference players. The big conference coefficient is more than 8 times as big as the (insignificantly different from zero) small conference player coefficient and indicates each collegiate point per minute corresponds to about 0.45 points per minute as a professional. A Chow test rejects pooling of big and small conference players, which means that the relationship between professional points per minute and collegiate points per minute and draft number are different for big and small conference players.

The lower panel of Table 7.1 shows the results for rebounds per minute. Again, the Chow test rejects pooling. More rebounds per minute in college implies more rebounds per minute as a professional, though the coefficient for a big conference player is about six times larger than that for a small conference player. For the big conference players, college rebounds are linked to about nine-tenths of a professional rebound. Both are significant at the 5% level or better. Interestingly, draft number is only significant, and at the 10% level, in the big conference regression. Oddly, the coefficient is positive, indicating that big conference players taken later in the draft collect more rebounds per minute than big conference players taken earlier in the draft. However, if one includes a dummy variable for centers in the equation, that variable is positive and significant at the 10% level (p = 0.05) while the draft position variable becomes insignificant though its sign remains positive.

Table 7.2 shows the assists per minute and steals per minute regressions. In neither case is draft number ever statistically significant nor is pooling of the small and big conference players rejected. Assists in college translate one for one into assists per minute as a professional while a collegiate steal per minute implies about a half a steal per minute in the NBA. Table 7.3 shows the relationship between collegiate blocked shots and professional blocked shots and between collegiate free throw percentage and professional free throw percentage. For blocked shots, draft position does not matter in either big or small conference or in the full sample. However, collegiate blocks per minute predicts professional blocks per minute in each subsample and in the full sample. The data reject pooling the samples, however, implying that an extra block per minute at a small conference school implies about .35 blocks per minute as a pro, but that same block per minute at a big conference school indicates .61 blocks per minute in the NBA.

Free throw shooting would seem likely to be a skill that transfers reasonably well from college to the NBA. The lower panel of Table 7.3 confirms this intuition. A Chow test cannot reject pooling small and big conference players. The full sample results indicate that players taken later in the draft are worse free throw shooters than those taken earlier. The estimated coefficient implies that the first player taken will shoot a bit under two percentage points better over his NBA career than the player taken 20th in the draft. The partial correlation between collegiate and professional free throw percentage is 0.789 and statistically significantly different from zero at the 1% level.

One might think that field goal shooting would translate well from college to the professional ranks, as did free throw shooting, and that better shooters would be drafted earlier, all other things constant. Early draftees do shoot better than later draftees, but the effect is small. The first player taken in the draft will only shoot about 1.8 percentage points better over his career than the 30th player taken. On the other hand, the results of Table 7.4 indicate that collegiate shooting does not predict professional shooting. The estimated coefficients have positive signs in all three models, and are, indeed, quite similar in size, but none is remotely statistically significant. The results indicate that pooling small and big conference players is not rejected by the data and that the typical NBA player will shoot about 33% from the field, all else constant. This result warrants further discussion.

Shots from the field involve defenders, and NBA defenders are surely better on average than defenders in college. In addition, the three-point shooting line is farther from the basket in the professional game than in college. For these reasons, shots taken in the NBA may be more difficult than shots taken in college, so the connection between college and professional shooting proficiency may be weaker. We began by checking for outliers and discovered that three observations had far lower NBA shooting accuracy than others. Omitting these three observations from the data, the results indicate that collegiate shooting percentage is a positive and statistically significant predictor of professional shooting percentage. The coefficient estimate is 0.29, so one percentage point of collegiate shooting accuracy translates into an additional 0.29 percentage points of professional shooting accuracy. Note that shooting from the field is much less predictable or consistent from college to the pros than is shooting from the free throw line, consistent with the better defenders and greater distance arguments above.

One other collegiate production variable that does well at predicting NBA career-long production is the turnover percentage. The lower panel of Table 7.4 shows that regardless of big or small conference, a player with a higher turnover percentage in college will have a statistically significantly higher turnover percentage during his career as a professional. The results indicate that one cannot reject the null that the coefficients are the same for the big and small conference players, indicating support for the full sample results. The coefficient on college turnover percent indicates for each one percentage point increase in turnover percent in college, the player will have a half a percentage point of turnover percent as a pro. In addition, not surprisingly, players drafted later have a higher turnover percentage. While the effect of draft position is statistically significant at the 10% level, it is small in practical terms. A player drafted 30 spots later than another will have about a one percentage point higher turnover percentage.

Table 8 reports estimates of the relationship between measures of productivity that combine multiple aspects of the game. We constructed the NBA productivity index and Berri, Schmidt, and Brook's (2006) Win Score for both a player's NBA career and his collegiate career. The results for the NBA productivity index indicate that big and small conference players can be pooled. In the pooled sample, draft number is statistically significant and negative, so being drafted later means a less productive career using the NBA metric. The collegiate productivity variable is not significant and has the wrong sign in the full sample regression. The collegiate productivity variable is, however, positive and statistically significant for players from big conference schools. The Win Score variable is also positive and significant for these players, though it is not significant in the pooled regression. Unlike the model using the NBA productivity variable, the Win Score regressions reject, at the 10% level, pooling of the small and big conference players. The draft position variable is not significant for the big conference players in explaining their professional Win Score, but it is negative and significant in the pooled regression. Neither Win Score nor draft position is significant in the small conference sample.

Discussion

The results in Tables 7.1 through 7.4 paint a clear picture that some types of NBA production over an entire career can be predicted fairly well based on a player's college production. These results are equally clear that not all types of professional production are predictable from college statistics. The results also indicate that frequently the correlation between college and professional productivity is different for players from big versus small conferences. Points per minute, rebounds per minute, blocks per minute, and turnover percentage are all larger in the NBA for a given level of the respective variable achieved in a big conference school than in a small conference school. Free throw percentage, assists per minute, and steals per minute all have nearly the same relationship for small and big conference players.

Perhaps more interesting still is the role of draft position in the relationship between college and professional production. After controlling for collegiate production per minute, draft position is no longer generally a statistically significant determinant of per minute production during the player's NBA career. In the points per minute, and turnover percentage equations, draft position is significant at the 10% level or better with the hypothesized sign, for at least some sample, small or big conference or the pooled sample. Draft position is positive and significant in the big conference rebounds per minute equation, but has the wrong sign, a result that disappears after controlling for position. Shooting, whether from the floor or from the free throw line, is statistically significantly better for players drafted earlier than for those drafted later, even after controlling for collegiate shooting proficiency. But the size of the impact of draft position is quite small. The implication is that once players are in the NBA, where they were drafted has little explanatory power for how productive they will be over their careers. Of the 24 draft position coefficients across the three samples and six performance statistics, only eight are statistically significant at the 10% level or better, and only two, both from the points per minute regressions, are significant at 1%. By contrast, 20 of the 24 coefficients on the college performance measures are significant at the 5% level or better. Interestingly, field goal shooting is the performance measure with the least transference from the college game to the pros, with the college variable never statistically significant as a predictor of professional accomplishment.

The bottom line is that there is a substantial degree of risk involved in drafting players, which the regressions in Tables 7.1 through 7.4 show. The R-squares are smaller than .7 in nearly every case, blocks and assists per minute among big conference players and assists per minute in the full sample being the exceptions, indicating that more than 30% of the variance in NBA productivity is unexplained by college production (and draft position). In most of the cases the unexplained portion of NBA production is more nearly 60 to 70% of production. In other words, drafting amateur players is a risky and uncertain endeavor. There will always be "can't miss" prospects that fail, and there will always be unheralded players that succeed. The goal of franchises is to have few of the former and many of the latter.

Consider again the questions posed at the start of the paper.

1. Can professional sports franchises identify quality players based on the information available to them prior to a draft? The answer for the NBA appears to be yes. Specific types of college productivity are significant determinants of draft position and generally significant predictors of NBA level production. However, there remains a great deal of variation in draft position and production as a professional that is unexplained by college productivity.

2. Do the attributes that determine draft position also indicate career success? As mentioned above, many college-level productivity measures are statistically significant predictors of NBA career productivity even after controlling for draft position. However, few college production variables are individually significant determinants of either whether a drafted player makes an NBA roster or the length of the career a drafted player will have. The best predictor of making a roster and length of career is draft position.

3. Do those attributes imply something about what the drafting teams value and are they the same attributes that professional teams compensate highly? Not really. The college-level statistics that are best at predicting NBA career productivity measures, as judged by R-squares, significant coefficients, and the size of the coefficients, are often not those that correlate most strongly with NBA compensation. For example, while the evidence in the literature is that scoring is the primary determinant of compensation in the NBA and is a significant determinant of draft position, at least for big conference players, college scoring is relatively weakly related to professional scoring. College rebounds, blocks, and assists have a much larger ability to predict professional output than does scoring, yet these are not generally found to be highly correlated with NBA salaries.

4. Are NBA economic decision makers rational? The evidence here suggests, in agreement with the published literature, that NBA teams may stick with early draft choices longer than late draft choices with the same productivity since early draft choices have longer careers, on average, than late draft choices. It also indicates that draft position is a significant determinant of playing time for players whose careers last five years or less but not for players whose careers are longer than five years. Whether this is irrational "escalation of commitment" or rational based on better information about expected costs and expected benefits cannot be determined from this analysis. Finally, there is some evidence here that NBA executives may draft players from small conferences in accord with the statistical discrimination and option value types of rationales from the literature.

Appendix

Table A1: Draft Position Regressions

VARIABLES (1) (2)
 Small Conference Big Conference

College Field Goal Pct. -2.263 ** -0.616
 (0.939) (0.500)
College Free Throw Pct. -1.613 ** -0.384
 (0.618) (0.335)
College points/minute 32.659 -78.290 ***
 (36.431) (26.894)
College rebounds/minute -10.841 -117.224 **
 (77.329) (55.837)
College assists/minute -58.240 -44.194
 (122.278) (78.217)
College steals/minute -296.375 -113.300
 (332.434) (168.004)
College blocks/minute -184.882 -216.549 **
 (202.694) (98.332)
College Turnover Pct. 1.081 -1.243
 (0.916) (0.821)
College fouls/minute 2.217 485.691 ***
 (196.470) (105.080)
height -1.410 -0.233
 (2.694) (1.067)
guard 5.406 -9.398
 (33.475) (10.808)
forward -0.937 -9.791
 (29.186) (8.001)
Constant 367.736 163.506 *
 (243.122) (94.331)
Observations 50 97
R-squared 0.374 0.499
Error Sum Squares 15100 21702

VARIABLES (3)
 Full Sample

College Field Goal Pct. -1.850 ***
 (0.429)
College Free Throw Pct. -0.603 *
 (0.305)
College points/minute -1.458
 (19.914)
College rebounds/minute 15.742
 (38.087)
College assists/minute -76.477
 (63.991)
College steals/minute -143.586
 (154.150)
College blocks/minute -150.099 *
 (89.697)
College Turnover Pct. 0.355
 (0.594)
College fouls/minute 301.724 ***
 (91.552)
height -1.715 *
 (1.015)
guard -9.097
 (10.979)
forward -12.385
 (8.292)
Constant 295.962 ***
 (89.477)
Observations 147
R-squared 0.322
Error Sum Squares 47109

*** p < 0.01, ** p < 0.05, * p < 0.1

Standard errors in parentheses

Table A2: Draft Position Regressions

VARIABLES (1) (2) (3)
 Small Conference Big Conference Full Sample

Collegiate NBA 5.718 -138.822 *** -22.315 **
 Prod. (12.714) (22.353) (9.294)
guard 35.809 * -19.651 ** -5.665
 (18.234) (8.198) (8.567)
forward 31.679 * -15.476 ** -7.600
 (17.502) (7.648) (8.040)
Constant 0.816 114.879 *** 49.068 ***
 (22.632) (14.057) (10.699)
Observations 52 97 149
Error Sum Squares 22114 29858 67060.640
R-squared 0.098 0.311 0.041
Prob > F 0.000
Chow test: F-stat 10.234

Standard errors in parentheses

*** p < 0.01, ** p < 0.05, * p < 0.1

Table A3: Draft Position Regressions

VARIABLES (1) (2) (3)
 Small Conference Big Conference Full Sample

Collegiate Win 8.544 -180.017 *** -38.843 **
 Score per minute (21.639) (33.209) (15.242)
guard 36.725 * -24.126 *** -7.395
 (19.327) (8.686) (8.813)
forward 30.678 * -15.753 ** -8.292
 (17.853) (7.932) (8.048)
Constant 2.377 83.376 *** 46.734 ***
 (21.613) (10.981) (9.722)
Observations 50 97 147
Error Sum Squares 21630 32099 66154.977
R-squared 0.104 0.260 0.047
Prob > F 0.000
Chow test: F-stat 8.037

Standard errors in parentheses

*** p < 0.01, ** p < 0.05, * p < 0.1

Authors' Note

This paper is an outgrowth of an independent study project by Babatunde Oguntimein under the supervision of Dennis Coates.

References

Basketball Reference. Retrieved from http://www.basketball-reference.com

Berri, D. J., Brook, S. L., & Schmidt, M. B. (2007). Does one simply need to score to score? International Journal of Sport Finance, 2(4), 190-205.

Berri, D. J., Schmidt, M. B., & Brook, S. L. (2006). The wages of wins: Taking measure of the many myths in modern sport. Palo Alto, CA: Stanford University Press.

Boulier, B. L., Stekler, H. O., Coburn, J., & Rankins, T. (2004). Evaluating National Football League draft choices: The passing game (Unpublished manuscript). Retrieved from http://www.umbc.edu/economics/seminar_papers/boulier_paper.pdf

Camerer, C. F., & Weber, R. A. (1999). The econometrics and behavioral economics of escalation of commitment: A re-examination of Staw and Hoang's NBA data. Journal of Economic Behavior and Organization, 39, 59-82.

Database basketball. Retrieved from http://www.databasebasketball.com

Groothuis, P. A., & Hill, J. R. (2004). Exit discrimination in the NBA: A duration analysis of career length. Economic Inquiry, 42(2), 341-349.

Groothuis, P. A., Hill, J. R., & Perri, T. (2007a). Early entry in the NBA draft: The influence of unraveling, human capital, and option value. Journal of Sports Economics, 8(3), 223-243.

Groothuis, P. A., Hill, J. R., & Perri, T. (2007b). The dilemma of choosing talent: Michael Jordans are hard to find. Appalachian State University Working Paper, 0701.

Hendricks, W., DeBrock, L., & Koenker, R. (2003). Uncertainty, hiring, and subsequent performance: The NFL draft. Journal of Labor Economics, 21(4), 857-886.

Kahn, L. M., & Sherer, P. D. (1988). Racial differences in professional basketball players' compensation. Journal of Labor Economics, 6(1), 40-61.

Massey, C., & Thaler, R. H. (2006). The losers' curse: Overconfidence vs. market efficiency in the National Football League. Retrieved from http://mba.yale.edu/faculty/pdf/massey_ thaler_overconfidence_nfl_draft.pdf

Spurr, S. J. (2000). The baseball draft: A study of the ability to find talent. Journal of Sports Economics, 1(1), 66-85.

Staw, B. M., & Hoang, H. (1995). Sunk costs in the NBA: Why draft order affects playing time and survival in professional basketball. Administrative Science Quarterly, 40(3), 474-494.

Winfree, J. A., & Molitor, C. J. (2007). The value of college: Drafted high school baseball players. Journal of Sports Economics, 8(4), 378-393.

Endnotes

(1) This approach differs from Staw and Hoang (1995), who use factor analysis to combine individual statistics into indices. Camerer and Weber (1999) use both the indices and the raw statistics.

(2) Because the number of rounds in the draft changes between the start and the end of our sample, we also estimated the model limiting the data to draftees from the first two rounds. The qualitative results are largely the same though college free throw percentage becomes significant in the big conference player draft position equation and personal fouls per minute is no longer significant in any equation. Interestingly, the elasticity of draft position with respect to collegiate free throw percentage is greater than any other elasticity in the big conference equation. The elasticity with respect to points per minute is the smallest among the significant variables, free throw percentage, rebounds, blocks, and assists. These results are available upon request.

(3) Groothuis, Hill, and Perri (2007a) suggested the option value argument applies for NBA draftees and assessed the possibility using draft of underclassman including players right out of high school. In our data very few underclassmen were drafted and all of them made the NBA. We do not have data on underclassmen that declared for the draft but were not selected.

(4) Neither Collegiate Win Score nor College NBA productivity is remotely significant when used in place of the vector of performance statistics. Results are available upon request.

Dennis Coates [1] and Babatunde Oguntimein [1]

[1] University of Maryland, Baltimore County

Dennis Coates is a professor in the Department of Economics. His research interests focus on the effects of stadiums and professional sports on local economies.

Babatunde Oguntimein is a graduate student in the John E. Walker Department of Economics at Clemson University. His research interests focus on the NBA and the evaluation of its players and coaches.

Table 1: Descriptive Statistics

Variable Small Conference

 N Mean SD

Draft Number *** 128 70.586 45.451
Ever played in NBA 128 0.469 0.501
Length of NBA Career 128 2.797 4.536
Big Conference
College Field Goal Pct. 55 51.395 5.011
College Free Throw Pct. 55 70.936 8.628
College points/minute 52 0.547 0.279
College rebounds/minute 52 0.223 0.155
College assists/minute 52 0.077 0.054
College steals/minute ** 52 0.042 0.022
College blocks/minute 52 0.031 0.035
College Turnover Pct. *** 52 13.679 4.936
College fouls/minute 52 0.087 0.02
College NBA Prod/ minute * 52 0.607 0.045
College Win Score/ minute 50 0.268 0.029
NBA Field Goal Pct. 60 42.229 8.112
NBA Free Throw Pct. 58 72.197 9.153
NBA points/minute 60 0.379 0.112
NBA rebounds/minute 60 0.171 0.088
NBA assists/minute 60 0.092 0.066
NBA steals/minute 60 0.037 0.017
NBA blocks/minute 60 0.018 0.02
NBA fouls/minute 60 0.119 0.052
NBA Prod/ minute 60 0.381 0.028
NBA Win Score/ minute 60 0.096 0.023

Variable Big Conference

 N Mean SD

Draft Number *** 162 51.488 41.097
Ever played in NBA 162 0.623 0.486
Length of NBA Career 162 4.642 5.549
Big Conference
College Field Goal Pct. 101 51.806 3.987
College Free Throw Pct. 101 72.114 7.725
College points/minute 97 0.463 0.088
College rebounds/minute 97 0.19 0.069
College assists/minute 97 0.078 0.054
College steals/minute ** 97 0.036 0.017
College blocks/minute 97 0.025 0.024
College Turnover Pct. *** 101 15.485 3.691
College fouls/minute 97 0.086 0.025
College NBA Prod/ minute * 97 0.517 0.009
College Win Score/ minute 97 0.216 0.006
NBA Field Goal Pct. 101 43.53 8.335
NBA Free Throw Pct. 100 70.481 15.422
NBA points/minute 101 0.38 0.089
NBA rebounds/minute 101 0.172 0.077
NBA assists/minute 101 0.089 0.064
NBA steals/minute 101 0.033 0.018
NBA blocks/minute 101 0.019 0.017
NBA fouls/minute 100 0.115 0.052
NBA Prod/ minute 101 0.402 0.011
NBA Win Score/ minute 100 0.121 0.009

Variable Full Sample

 N Mean SD

Draft Number *** 290 59.917 44.034
Ever played in NBA 290 0.555 0.498
Length of NBA Career 290 3.828 5.2
Big Conference 290 0.559 0.497
College Field Goal Pct. 156 51.661 4.364
College Free Throw Pct. 156 71.699 8.047
College points/minute 149 0.493 0.183
College rebounds/minute 149 0.202 0.108
College assists/minute 149 0.078 0.054
College steals/minute ** 149 0.038 0.019
College blocks/minute 149 0.027 0.028
College Turnover Pct. *** 153 14.871 4.227
College fouls/minute 149 0.086 0.023
College NBA Prod/ minute * 149 0.548 0.017
College Win Score/ minute 147 0.234 0.011
NBA Field Goal Pct. 161 43.045 8.251
NBA Free Throw Pct. 158 71.111 13.456
NBA points/minute 161 0.38 0.098
NBA rebounds/minute 161 0.171 0.081
NBA assists/minute 161 0.09 0.064
NBA steals/minute 161 0.035 0.018
NBA blocks/minute 161 0.018 0.018
NBA fouls/minute 160 0.116 0.052
NBA Prod/ minute 161 0.394 0.013
NBA Win Score/ minute 160 0.112 0.010

(a) Big conferences: Big 8, Big 10, Pac 10, ACC,
Southeast, Southwest, Big East, Metro

*** p < 0.01 ** p < 0.05 * p < 0.10

Table 2: Draft Position Regressions

VARIABLES (1) (2) (3)
 Small Big Full Sample
 Conference Conference

College Field Goal Pct. -2.269 ** -0.557 -1.823 ***
 (0.859) (0.493) (0.432)
 [-3.31] [-3.13]
College Free Throw Pct. -1.664 *** -0.454 -0.673 **
 (0.582) (0.326) (0.306)
 [-3.35] [-1.60]
College points/minute 45.053 -76.981 *** 11.299
 (30.149) (26.412) (19.342)
 [-1.30]
College rebounds/minute -55.522 -128.676 ** -20.798
 (54.373) (50.457) (34.882)
 [-0.89]
College assists/minute -45.504 -28.510 -38.646
 (108.034) (74.568) (61.655)
College steals/minute -120.041 -136.375 -24.067
 (231.679) (153.650) (137.254)
College blocks/minute -194.119 -198.898 ** -145.907 *
 (146.163) (89.376) (81.560)
 [-0.18] [-0.13]
College Turnover Pct. 1.233 -1.301 0.421
 (0.874) (0.806) (0.598)
College fouls/minute 3.650 503.717 *** 283.734 ***
 (175.989) (101.741) (89.775)
 [1.58] [0.82]
Constant 254.879 *** 138.103 *** 148.095 ***
 (71.263) (40.541) (37.769)
Observations 50 97 147
R-squared 0.357 0.490 0.292
Error Sum Squares 15514 22105 49180
Chow test: F-stat 3.903
Prob > F 0.000

Standard errors in parentheses

Elasticities in brackets

*** p < 0.01, ** p < 0.05, * p < 0.1

Table 3: Draft Position Regressions

VARIABLES (1) (2) (3)
 No Championship Full Sample
 Championship Game
 Game

College Field Goal Pct. -1.835 *** 0.367 -1.823 ***
 (0.466) (0.745) (0.432)
College Free Throw Pct. -0.939 *** -1.002 * -0.673 **
 (0.334) (0.507) (0.306)
College points/minute 12.545 -116.853 ** 11.299
 (19.984) (44.895) (19.342)
College rebounds/minute -12.371 -251.440 *** -20.798
 (35.896) (76.942) (34.882)
College assists/minute -58.034 155.975 -38.646
 (66.399) (107.090) (61.655)
College steals/minute 47.919 -254.051 -24.067
 (146.083) (223.741) (137.254)
College blocks/minute -153.509 * -179.093 -145.907 *
 (83.931) (162.460) (81.560)
College Turnover Pct. 0.699 -4.966 *** 0.421
 (0.608) (1.260) (0.598)
College fouls/minute 51.072 1,045.083 *** 283.734 ***
 (105.620) (154.106) (89.775)
Constant 180.943 *** 164.962 *** 148.095 ***
 (41.153) (58.487) (37.769)
Observations 115 32 147
R-squared 0.251 0.836 0.292
Error Sum Squares 33278 4092 49180.211
Chow test: F-stat 4.014
Prob > F 0.000

Standard errors in parentheses

*** p < 0.01, ** p < 0.05, * p < 0.1

Table 4: Career Length Regressions-All Draftees

VARIABLES Probit Tobit Negative binomial

Draft Number -0.071 *** -0.258 *** -0.024 ***
 (0.015) (0.040) (0.009)
Draft Number Squared 0.00026 *** 0.001 *** -0.00004
 (0.00009) (0.000) (0.00006)
Big Conference * -0.006 -0.033 -0.013 **
 Draft Position (0.007) (0.025) (0.006)
Big Conference 0.387 1.679 0.461 *
 (0.493) (1.305) (0.278)
Constant 3.161 *** 11.595 *** 2.302 ***
 (0.573) (1.303) (0.270)
Alpha 1.068 ***
 (0.156)
Observations 290 290 290

*** p < 0.01, ** p < 0.05, * p < 0.1

Standard errors in parentheses

Table 5: Career Length

VARIABLES Tobit Negative binomial

Draft Number -0.281 *** -0.040 ***
 (0.059) (0.010)
Draft Number Squared 0.002 *** 0.00031 ***
 (0.001) (0.00011)
Big Conference * Draft Position -0.017 -0.002
 (0.037) (0.006)
Big Conference 1.582 0.188
 (1.439) (0.236)
College Field Goal Pct. 0.097 0.007
 (0.103) (0.017)
College Free Throw Pct. -0.167 ** -0.029 **
 (0.071) (0.012)
College points/minute 1.339 0.142
 (4.471) (0.755)
College rebounds/minute -4.983 -1.146
 (8.508) (1.448)
College assists/minute 21.695 2.130
 (14.508) (2.637)
College steals/minute -5.903 0.269
 (34.542) (5.821)
College blocks/minute -18.036 -3.274
 (20.600) (3.533)
College Turnover Pct. -0.145 -0.020
 (0.137) (0.026)
College fouls/minute -24.667 -3.657
 (22.654) (3.947)
height 0.236 0.041
 (0.229) (0.035)
guard 4.849* 0.769*
 (2.539) (0.427)
forward 3.014 0.495
 (1.941) (0.341)
Constant 0.157 1.190
 (21.057) (3.271)
Alpha 0.331 ***
 (0.062)
Observations 147 147
Wald test 1.228 3.467
Prob > 0 0.302 0.325

*** p < 0.01, ** p < 0.05, * p < 0.1

Standard errors in parentheses

Table 6: Minutes Played Regressions

VARIABLES (1) (2) (3)
 Career < 6 Career > 5 Full Sample

College Minutes per game 0.195 * 0.313 * 0.222
 (0.111) (0.181) (0.135)
Guard 5.618 ** 2.675 9.832 ***
 (2.579) (3.904) (2.980)
Forward 4.291 * 0.282 5.730 **
 (2.294) (3.615) (2.737)
Draft Number -0.274 *** -0.197 -0.568 ***
 (0.096) (0.123) (0.081)
Draft Number Squared 0.002 ** 0.001 0.004 ***
 (0.001) (0.002) (0.001)
Constant 6.652 * 18.056 *** 15.673 ***
 (3.768) (6.025) (4.404)
Observations 75 74 149
R-squared 0.203 0.289 0.403
Error Sum Squares 1687.574 2292.590 7911.072
Chow test: F-stat 22.551
Prob > F 0.000

*** p < 0.01, ** p < 0.05, * p < 0.1

Standard errors in parentheses

Table 7.1: Points and Rebounds Regressions
Points per minute
 (1) (2) (3)
VARIABLES Small Big Full Sample
 Conference Conference

Draft Number -0.00198 *** -0.00048 -0.00147 ***
 (0.00063) (0.00041) (0.00033)
College points/minute 0.052 0.443 *** 0.095 **
 (0.050) (0.099) (0.039)
Constant 0.415 *** 0.192 *** 0.378 ***
 (0.040) (0.053) (0.024)
Observations 52 97 149
Error Sum Squares 0.473 0.520 1.079
R-squared 0.195 0.270 0.172
Chow test: F-stat 4.141
Prob > F 0.008

Rebounds per minute

Draft Number -0.00040 0.00040 * 0.00004
 (0.00053) (0.00021) (0.00027)
College rebounds/minute 0.156 ** 0.908 *** 0.353 ***
 (0.075) (0.065) (0.054)
Constant 0.150 *** -0.010 0.100 ***
 (0.029) (0.015) (0.015)
Observations 52 97 149
R-squared 0.096 0.677 0.230
Error Sum Squares 0.337 0.179 0.716
Chow test: F-stat 18.343
Prob > F 0.000

Standard errors in parentheses

*** p < 0.01, ** p < 0.05, * p < 0.1

Table 7.2: Assists and Steals Regressions
Assists per minute

VARIABLES (1) (2) (3)
 Small Big Full Sample
 Conference Conference

Draft Number -0.00003 -0.00003 -0.00001
 (0.00026) (0.00013) (0.00012)
College assists/minute 1.014 *** 1.080 *** 1.057 ***
 (0.107) (0.053) (0.050)
Constant 0.017 0.004 0.008
 (0.013) (0.006) (0.006)
Observations 52 97 149
Error Sum Squares 0.0795 0.0733 0.155
R-squared 0.656 0.815 0.754
Chow test: F-stat 0.721
Prob > F 0.541

Steals per minute

Draft Number -0.00009 -0.00003 -0.00005
 (0.00007) (0.00008) (0.00006)
College steals/minute 0.587 *** 0.454 *** 0.508 ***
 (0.067) (0.099) (0.063)
Constant 0.014 *** 0.018 *** 0.016 ***
 (0.004) (0.005) (0.003)
Observations 52 97 149
R-squared 0.612 0.188 0.308
Error Sum Squares 0.00524 0.0261 0.032
Chow test: F-stat 0.486
Prob > F 0.693

Standard errors in parentheses

*** p < 0.01, ** p < 0.05, * p < 0.1

Table 7.3: Blocks and Free Throw Percentage Regressions

Blocks per minute

VARIABLES (1) (2) (3)
 Small Big Full Sample
 Conference Conference

Draft Number 0.00002 0.00001 0.00001
 (0.00011) (0.00004) (0.00005)
College blocks/minute 0.354 *** 0.606 *** 0.465 ***
 (0.069) (0.037) (0.037)
Constant 0.006 0.003 * 0.006 **
 (0.005) (0.002) (0.002)
Observations 52 97 149
Error Sum Squares 0.0135 0.00720 0.023
R-squared 0.360 0.737 0.524
Chow test: F-stat 5.543
Prob > F 0.001

Free Throw Percentage

Draft Number -0.06831 * -0.14173 ** -0.09962 **
 (0.04023) (0.06620) (0.04310)
College Free Throw Pct. 0.732 *** 0.829 *** 0.789 ***
 (0.114) (0.178) (0.120)
Constant 22.441 ** 14.617 17.452 *
 (8.393) (13.333) (8.944)
Observations 54 100 154
Error Sum Squares 2485 18005 20980.104
R-squared 0.476 0.235 0.261
Prob > F 0.320
Chow test: F-stat 1.180

*** p < 0.01, ** p < 0.05, * p < 0.1

Standard errors in parentheses

Table 7.4: Field Goal and Turnover Percentage Regressions

Field Goal Percentage
 (1) (2) (3)
VARIABLES Small Big Full Sample
 Conference Conference

Draft Number -0.07578 -0.047 -0.06314 **
 (0.04763) (0.042) (0.03085)
College Field Goal Pct. 0.205 0.242 0.223
 (0.227) (0.220) (0.157)
Constant 34.060 *** 32.294 *** 33.359 ***
 (12.330) (11.857) (8.516)
Observations 55 101 156
R-squared 0.082 0.036 0.056
Error Sum Squares 3280 6697 10031.440
Chow test: F-stat 0.272
Prob > F 0.845

Turnover Percentage

Draft Number 0.03805 0.02517 0.03323 *
 (0.02633) (0.02341) (0.01725)
College Turnover Pct. 0.309 ** 0.686 *** 0.492 ***
 (0.130) (0.133) (0.091)
Constant 9.165 *** 3.593 * 6.527 ***
 (1.883) (2.050) (1.403)
Observations 52 101 153
R-squared 0.173 0.250 0.205
Error Sum Squares 954.9 2205 3254.485
Chow test: F-stat 1.474
Prob > F 0.224

*** p < 0.01, ** p < 0.05, * p < 0.1
Standard errors in parentheses

Table 8: NBA Production Index Regressions

 NBA Index

VARIABLES (1) (2) (3)
 Small Conference Big Conference Full Sample

Draft Number -0.003 * -0.001 -0.002 ***
 (0.001) (0.001) (0.001)
College NBA -0.035 0.293 ** -0.013
 Prod/minute (0.094) (0.138) (0.060)
Constant 0.492 *** 0.277 *** 0.461 ***
 (0.086) (0.081) (0.042)
Observations 52 97 149
R-squared 0.069 0.132 0.074
Error Sum Squares 2.260 0.889 3.222
Prob>F 0.350
Chow test: F-stat 1.102

Berri, Schmidt, and Brook's Win Score Index Regressions

Draft Number -0.002 -0.001 -0.002 ***
 (0.001) (0.001) (0.001)
College Win -0.112 0.507 *** -0.042
 Score/minute (0.132) (0.146) (0.082)
Constant 0.191 *** 0.030 0.169 ***
 (0.065) (0.039) (0.029)
Observations 50 96 146
R-squared 0.058 0.195 0.065
Error Sum Squares 1.585 0.590 2.287
Chow test: F-stat 2.405
Prob>F 0.070

Standard errors in parentheses

*** p < 0.01, ** p < 0.05, * p < 0.1