Assessing the Relative Value of Draft Position
in the NBA Draft

Introduction

Success in the draft is crucial to the success of an NBA team. With their accomplishments in the last decade, the San Antonio Spurs are often cited as an example of the draft’s importance. The Spurs had the good fortune to win the draft lottery in 1997, which allowed them to draft Tim Duncan. Later, they demonstrated their skill by drafting Manu Ginobili and Tony Parker with the 57^th pick in 1999 and 28^th in 2001.

As advanced quantitative analysis has become more prominent in the NBA generally of late, it has also become more prevalent in draft preparation. A variety of quantitative approaches to the draft have been discussed in the media. These include the Trail Blazers arrangement with Jeff Ma and ProTrade, John Hollinger’s new system for rating draft-eligible players, and Ed Weiland’s articles for HoopsAnalyst.com.

While these efforts aim to rate the value of players, they don’t attempt to quantify the value of holding the rights to a particular pick in the draft. This is an important aspect of the draft that is relevant both on draft night and throughout the season, as many trades involve the exchange of future draft picks. It is particularly important the night of the draft lottery, but attempts to quantify the expected value of moving up to the 1^st pick from the 6^th pick as Portland did this year are rarely mentioned. Similarly, while there was a recognition that exchanging 2007 draft picks as part of the Eddy Curry trade aided Chicago, there was limited discussion of how much more value Chicago could expect to obtain with the 9^th pick rather than the 23^rd once the season concluded.

This article aims to quantify the relative value of picking in different positions in the NBA draft. This topic appears to be relatively neglected in publicly available analysis, with Kevin Pelton’s interesting 2003 article serving as the primary public source of information. Kevin assessed the draft from 1995 through 2001 using his Value Over Replacement Player formula as the metric to value players. In this article, the results of the draft from 1980 through 2003 are considered, resulting in up to 24 observations for a given draft position. Additionally, multiple metrics are considered in this analysis in an attempt to develop a "metric-neutral" assessment of value, one that both a proponent of PER and a Win Shares evangelist would both be comfortable using.

The data used in this analysis was graciously provided by Justin Kubatko of basketball-reference.com. The dataset included the draft results from 1980-2006, the advanced performance data for each player for the 1980-2007 regular seasons, and the estimated salaries for each player from 1980 through 2007. Without Justin’s database of information, this analysis would not be possible.

As stated above, a key goal of this analysis was to develop an assessment that isn’t overly dependent on one particular metric. To ensure the results would be meaningful to a wide audience that may prefer one advanced statistic over another, values were assessed using four statistics:

• PER-Minutes (Season PER*Season Minutes)
• Player Wins
• Win Shares
• Estimated Salary

These metrics are all defined in the basketball-reference.com glossary. PER, developed by John Hollinger, is perhaps the most widely used "advanced" NBA statistic. It is a per-minute measure of player efficiency, so it has been multiplied by minutes played to assess the cumulative value a player contributed over the course of a season in terms of PER. Player Wins and Win Shares, as reported by basketball-reference.com, are both cumulative statistics that estimate how many wins (or shares of a win) are attributable to a player over the course of a season. Estimated Salary is also considered in this analysis as it is a market assessment of the value of a player, though it is distorted by a variety of NBA Salary Cap rules (e.g. the mid-level exception, rookie salary scale).

In addition to exploring different statistics, the cumulative value of each statistic was assessed for multiple time periods. The three time periods were:

• Career
• First 4 Years
• Years with Rookie Team

Evaluating a player over their entire career is probably the most straightforward and intuitive method to assess the value of a player. However, all of the value in a player’s career is not always realized by the team that drafted him. Therefore, players were also evaluated on the value generated in their first 4 seasons after the draft as well as with their rookie team. The first 4 years is a relevant time period because under the current collective bargaining agreement rookies sign a two year contract with a team option for the third and fourth year. After the fourth year, a player becomes a restricted free agent and has slightly more control over where they play. Other time periods could have been chosen, such as the minimum number of years that a talented NBA player must play for their drafting team before becoming an unrestricted free agent (5 years), but the clear definition of restricted free agency eligibility is preferred to the variability in a player’s eligibility for unrestricted free agency.

Additionally, the value a player delivered to their rookie team over the course of their career was calculated. This was hypothesized to be a more accurate assessment of value obtained through the draft as it measures the value a player delivered for their rookie team on the court. The rookie team is used rather than drafting team to account for draft day trades (e.g. Kobe Bryant, Vince Carter). The on-court value captures a significant fraction of a player’s total value to a team, but when a player leaves a team he can add value through a trade or freeing up salary cap space. Ideally both components would be considered, but the logistics of assessing this "departure" value is both complicated and ambiguous in the case of multi-player trades.

Each drafted player’s value (2,698 players) was determined based on these 12 metrics (4 statistics for three time periods). In practice, estimated salary was considered to be meaningful only over the time frame of a career due to the correlation of salary with draft position early in a player’s career, leaving 10 metrics for consideration. The relative value of each player compared to the top draft choice of their draft was then calculated for each of the metrics. By assessing relative value instead of absolute value, variations in the league over time are addressed in general, such as changes in scoring or salary as well as the 1998-99 lockout-shortened season. However, these variations are not completely accounted for as player’s careers are of varying length.

Figure 1 illustrates a sample result for one of the 10 metrics, Career PER-Minutes. The figure plots the average value of each draft position relative to the first pick, and includes the 95% confidence interval for each point. The sizable confidence intervals indicate that there is a fair bit of variability in the value of drafting in a given position from year to year. This is likely the result of a number of factors, particularly the strength of a draft, the accuracy of conventional wisdom, and the skill of the general managers responsible for making the selection. It is interesting to see that the second pick has yielded less value on average, in terms of Career PER-Minutes, than the third, fourth, and fifth pick, though this difference is not statistically significant.

A clear trend of exponential decay is evident in Figure 1 despite the confidence intervals. The best fit exponential curve, also illustrated in the figure, is expressed as

(1)

The quality of the fit is quite good as measured by R², the coefficient of determination. R² represents the proportion of the variation in the data that is accounted for by the model.

For this metric, the exponential curve fits the data quite well with an R² of 0.87, indicating that 87% of the change in relative value as draft position increases is explained by the model.

Figure 2 shows the data and best fit exponential curve for 4 Year PER-Minutes. Again, the second pick has yielded less value than the third on average using this metric. The best fit curve for is given by 4 Year PER-Minutes

(2)

and has an R² of 0.91. The results for 4 Year PER-Minutes are slightly better than those for Career PER-Minutes, with smaller confidence intervals and an improved fit. However, Figure 3 illustrates that the two estimates of value are quite similar, as expected given the similar coefficients.

Figure 4 shows the data and best fit exponential curve for Rookie Team PER-Minutes. The data exhibits a much larger degree of variability for the rookie team time frame than for either the career or 4 year time frame. The best fit exponential curve is a much poorer fit with an R² of 0.64 and a curve that indicates the second pick has a higher value than the first. There appear to be significant outliers in the data with much higher values than would be expected from a draft position. This is likely a result of players in these positions having exceptionally long careers with their rookie teams.

The results shown in Figures 1 to 4 are typical for the other statistics, with the rookie team data showing more variability and less plausible results than the career or 4 year data. As a result, the remaining analysis focused on the career and 4 year time frames. Curves for all time frames for Player Wins, Win Shares, and Estimated Salary, along with a comparison for the two time frames for each statistic are shown in the Appendix.

Figures 5 and 6 illustrate the similarity of estimates of value using different statistics for a given time frame. The results for PER-Minutes and Win Shares are strikingly similar, with Player Wins indicating more value for later draft picks than the aforementioned metrics and Estimated Salary less value for later picks. The two Player Wins metrics have the lowest R² of the seven metrics of interest with a value of 0.77 for Career Player Wins and 0.81 for 4 Year Player Wins, while Career Estimated Salary has the second highest R² with a value of 0.89. Table 1 summarizes the best fit lines and R² for the seven metrics of interest.

Table 1: Best fit curve for the seven metrics of interest