By Diego Arguelles
Whether you’re a team in the perpetual doldrums, a once storied franchise desperately looking to regain past glory, or a steady force wanting to keep your place atop the NBA’s Mount Olympus, drafting a 2nd round pick that exceeds their value can go a long way in helping achieve those objectives. If done correctly, a team can draft a player that far exceeds not only their pick value but also provides tremendous economic value. NBA rookies are restricted to scaled contracts under the NBA’s collective bargaining agreement (CBA) where the league designates yearly salary based on what pick that player is. Players drafted in the 2nd round have the potential to be especially valuable commodities given the fact that they make less money than a 1st round pick and potentially, a team can sign a 2nd rounder for 3 or 4 years at less than $1 million per year. These multi-year, cheap contracts for 2nd round picks have become so valuable that if one of these 2nd rounders turns out to be a significant contributor, it can feel like hitting the NBA asset jackpot. The most salient current example would be Draymond Green; an undervalued “tweener” who teams passed on all the way to the 35th pick in the 2012 Draft. Now, Green is an All-Star and by many accounts, one of the best 15 players in the league. Green’s contract flexibility, before re-signing with the Warriors in 2015, is a prime example of how important these players can become.
All of this sounds terrific for NBA front offices, but there is one problem; 2nd round picks that contribute at that level, or even at all, are extremely rare. These 2nd round “steals” are perpetually in higher demand than they are actually available. That being said, with the right combination of film work, personal evaluation, and certain statistical analyses, they can be found.
Given the amount of statistics we have at our disposal, I wanted to see if it was possible to analyze certain data sets from these 2nd rounders and uncover any possible leads to finding these once hidden gems. Or I’d just waste my time combing through the minutiae of Mike Muscala’s collegiate stats from the Patriot League. Either one. After taking the 150 2nd round picks that have been drafted in the 5 drafts from 2010-2014, along with the use of statistical models based on a criteria of 2nd round picks’ “success” in the NBA, there does seem to be some level of statistical correlation between the college stats of successful 2nd round picks and non-successful 2nd round picks when looking at 5 key benchmarks: Player Efficiency Rating (PER), Offensive Win Shares (OWS), Defensive Win Shares (DWS), Defensive Box Plus-Minus (DBPM), and Defensive Rating (DRtg).
Attributes of a “Steal”
First of all, determining what classifies as a “successful” 2nd round draft pick was necessary in order to juxtapose that group of players from the rest of the 2nd rounders that did not, or have not, become successes. I also wanted to come up with the most objective parameters possible in order to determine a draftee’s success so that the results are substantiated, even with the small sample size. In order to maintain this objectivity, I classified successful 2nd round picks, or steals, in two ways. The first way to earn the moniker of steal would be for players to have produced at least 1.0 wins for their respective teams (win shares) as well as having played at least 5 seasons in the NBA or every season in the NBA since joining the league. The second way, or the Hassan Whiteside Method, would be if a player has had a career average PER that equals, or surpasses, 15.0.
A 2nd round pick contributing 1.o wins to their team might seem low but if you look at the last 5 years, not many 2nd round picks have actually done that and finding a player in the 2nd round that can contribute directly to a win should be considered a success, given the dearth of impact players that historically come from the 2nd round. The reason there’s the added distinction of years played is due to the average career length in the NBA. According to Larry Coon, CBA expert and general NBA virtuoso, the average NBA career length is 4.8 years. Because the odds of drafting a player in the 2nd round that will go on to have that level of career longevity are so low, making the criteria 5 years would indicate the player has had an above-average tenure length in the NBA. However, since we are working with a sample size that is only 5 years old, players who were drafted after 2010 obviously haven’t had the chance to play 5 years yet but playing every season since joining the NBA should count for their career longevity. This also includes those who started overseas and have played every season in the NBA since their arrival. The Hassan Whiteside Method simply allows for a player to bounce around the league (or world) and still be considered a steal if they’ve played well enough. “Well enough”, in this case, is averaging a 15.0 PER for your career, which basically means you have played average-to-good basketball during that time. For general reference, a 15.0 PER is about an average NBA player’s season, 20.0 is usually an All-Star caliber season, 25.0 is an MVP caliber season, and 30.0+ is a holy-s#^*-has-only-happpened-19-times-ever season, per Basketball Reference. Since his return to the NBA, Whiteside has had a PER of 23.8 in 2014-2015 and 29.1 in 2015-2016 with the Heat.
In order to have the most accurate data possible, data collection of college basketball stats took place on Sports-Reference.com by way of Basketball-Reference.com, exclusively. There was no statistical database that had the breadth of traditional, advanced, and per 100 possession statistics that Basketball Reference did. In order to keep consistency, data was collected for every 2nd round pick from 2010-2014 that played college basketball. Because of the lack of advanced statistics for some of the International 2nd round picks, including these players would have thrown off the consistency of the data sets. Unfortunately, this excludes players such as Nikola Jokic and Bojan Bogdonavic, who are considered 2nd round steals under the established parameters. All 2nd round picks that had college basketball statistics and that were drafted from 2010-2014 had their data collected. Collection included each of their collegiate career’s traditional statistics, advanced statistics, and each player’s career ORtg and DRtg. These 44 different metrics were collected and if a player’s NBA win shares (1.0) combined with career length, or 15.0 career PER, indicated that they were a success, they were given a “1” while those who were not classified as NBA successes were given a “0”.
The Diamonds within the Rough, 2010-2014
Once the 150 players had their statistics collected, 37 players were excluded due to a lack of collegiate statistics (overseas or D-League professionals). This finalized the data set of 113 players with collegiate statistics from the 2010-2014 drafts. Out of these 113 players drafted, 29 of them (25.66%) were considered successes. These players have had varying levels of success at the NBA level but their play has exceeded their draft positions nonetheless.
The binary classification of all collected players (successful group=“1” and non-successful group=“0”) was then run through the models. Once the Spearman Correlation (with the binary variables) was run through the model, the initial results showed that the relationships between collegiate PER (.191*), OWS (.198*), DWS (.225*), DBPM (.225*), DRtg (-.209*) and those players’ future NBA success are statistically significant. However, the magnitudes of these values indicated that these factors have somewhat weak relationships.
A logistic regression was then performed to further test the effects that PER, OWS, DWS, DBPM, and DRtg have on success in order to find a stronger statistical significance. In the linear regression, R2 represented the proportion of variance of the outcome that can be explained by the predictors. However, for the logistic regressions and ordinal regression models, since it’s impossible to compute the R2 statistic, the other approximations were computed instead. These pseudo R2 values (e.g., Nagelkerke R Square=22.0%) indicate that the effect on the five factors account for a relatively low acceptable proportion of the variation on the success of the players. The results indicated that the five factors were not statistically significant at the 0.05 level. This might have been due to the unbalanced sample size combined with other external factors (e.g., player positions, years played) that could have impacted these results.
In order to keep trying to find the most significantly correlated differences between the successful group and the non-successful group, a T-Test was performed. This T-Test compared the mean differences between the groups and the results showed that the means of PER (t=-2.12, p < .05), DWS (t=-1.71, p <.05), DBPM (t=-2.251, p <.05), and DRtg (t=2.14, p <.05) were, in fact, statistically different. The important aspect to note from this test is that the successful group has a higher mean value than the non-successful group in PER (23.13 vs. 21.46), DWS (6.05 vs. 4.98), and DBPM (3.43 vs. 2.33), while having a lower value than the non-successful group in DRtg (95.25 vs. 97.47), which further helps prove a correlation since the lower the DRtg, the better that player is rated on defense.
These results mean that the average of the successful group’s PER, DWS, DBPM, and DRtg are all significantly different than the average of the non-successful group, somewhat validating the general assessment that certain metrics could help predict more successful 2nd round picks, relative to their less successful peers. To take it a step further and including OWS, according to these models, we can reasonably imply that a player drafted in the 2nd round from 2010-2014, with a collegiate PER >23.13, OWS >8.05, DWS >6.05, DBPM >3.43, and a DRtg <95.25, would have a significantly higher chance to be “successful” in the NBA, by our criteria, than a player who didn’t meet any (or as many) of these benchmarks. There was a significant enough difference, given the sample size constraints, between the successful and non-successful groups that it was appropriate to set these thresholds of the 5 different T-Test means. Additionally, after going back through the entire data set of both groups, there was only one false-positive that met all 5 of the T-Test benchmarks. That player is Arsalan Kazemi who actually played professional basketball before his collegiate career at Rice University for 3 years and then the University of Oregon for 1 year. One false-positive in the entire rest of the data set of 113 picks is an encouraging enough sign.
Finding Draft Foresight
While the sample size is small, given the fact that there are only 30 2nd round picks per year, there is still a relative validity to these trends, especially after the final T-Test. The fact that there was a statistically significant difference in the means of PER, DWS, DBPM, and DRtg not only shows that these advanced statistics can be helpful indicators of 2nd round success, it also shows how important advanced defensive metrics are to 2nd round prospects. Most of these draft steals rated out highly in DWS, DBPM, and/or DRtg so while the excitement around some younger 1st round picks may be geared more towards raw potential and offensive output, many of these more experienced 2nd round gems have become successful in the NBA due in large part to their defensive acumen. In terms of future application, these tests can be applied to future NBA Drafts to see if these results hold and to predict possible steals before they’re drafted. Many factors go into a player’s NBA success apart from these statistics so by no means is this, or any other singular model, the smoking gun but it would be fascinating to see which players from future drafts fit into the successful statistical correlation of PER, OWS, DWS, DBPM, and DRtg and follow their careers to see if these trends hold.
Undoubtedly, there will always be a place for film study, interviews, and personal evaluations in the NBA Draft process. In fact, those are some of the most vital pieces of research material that go a long way in determining which players will be most successful in the NBA. That being said, when you get to the 2nd round, front offices must use every resource at their disposal in order to find that elusive, and cost controlled, steal. The statistical tests that were conducted, while by no means iron clad, can be a valuable asset to a sports organization looking to get the slightest of edges in finding that diamond in the rough. Although further fine tuning could yield even more significant correlations, these current tests have found statistically significant factors that were present in a high majority of 2nd round steals of the last 5 years.
To put these statistically significant factors into perspective from another point of view, using our 5 main T-Test benchmarks of PER >23.13, OWS >8.05, DWS >6.05, DBPM >3.43, and a DRtg <95.25, out of the the 29 players in the successful group, 26/29, or 89.66%, of the 2nd round steals of the last 5 years met at least one of these benchmarks. 14/29, or 48.28%, met at least three out of the five benchmarks. Most notably, two 2nd round steals from the last 5 years met all five of the statistically significant T-Test benchmarks. One of them is Mike Muscala, who was a key figure for Bucknell in the relatively small Patriot League, possibly contributing to the gaudy collegiate statistics he put up. The other player is the paragon of 2nd round steals. The player that all current NBA general managers wish they would’ve taken (and claim that they would have if not for a, b, and c); two-time Defensive Player of the Year runner-up, 2016 All-Star, and NBA Champion, Draymond Green.
While this of course does not guarantee finding a player of Green’s caliber in the draft, along with traditional film work, interviews, and player evaluations, using statistical models that find correlated patterns of success, such as the 5 T-Test benchmarks found here, can help an NBA front office find an edge where others don’t.