By: Kiel Messinger and Zayne Kratz
Introduction
Draft picks are as important as ever in today’s NBA. From the Sam Hinkie 76ers introducing the idea of “tanking” to mainstream NBA media under the name of “The Process,” to the Thunder collecting draft picks like they’re Anthony Davis injuries, teams are building more and more for the future using these lottery tickets. But how valuable are these tickets, and what are their chances of cashing at each pick? How much are picks really worth in relation to each other, and players in the league?
To start, we have to define worth on the basketball court. For this, we are going to use John Hollinger’s Player Efficiency Rating, which gives each player an overall rating using a number of statistical measurements. Before moving on, I will acknowledge that this rating system is not perfect; Hollinger himself has pointed out flaws like undervaluing lockdown defensive players, and from my analysis it over values certain rebounding big men (you can’t tell me that Deandre Ayton is more valuable this year than Steph Curry and Jarrett Allen is more valuable than Donovan Mitchell), but it does give a basic sense of overall performances and value to a team.
This year’s top 5 so far are:
Nikola Jokic (32.93)
Giannis Antetokoumpo (32.64)
Joel Embiid (31.70)
LeBron James (26.32)
Kevin Durant (25.79)
… and it’s hard to argue with that. It is important to note that the league average PER is 15, and the following table which briefly indicates an estimation of what a rating means over a season.
Based on this table, and adjusting for the fact that most of our samples are of multiple-year spans which include injuries, we’ve made our own player-defining rating tier system using PER values.
Using the PER data from 2002-2020, we can get a sense of the value of picks as a result of the PER of the players they have produced over a stretch. We can also get insight into what picks are worth in relation to players over a specified amount of time.
Data and Analysis
To analyze pick performance, we are going to use their PER from years 4 to 6, which gives them enough time in the league to reach their potential while giving us enough years of data (2001-2015). The following graph shows a player’s mean PER over these years of their career, with each point representing a player. The lighter the color of the point, the more recent of a draft pick they are.
The graphs show the expected downward trend in terms of PER as the pick gets worse. The graph had a slope of -.315 and y-intercept of 15.40, indicating that according to the linear model, the first pick would be worth 15.40 PER over the three year stretch, and each resulting pick would be worth .315 less. The R-squared is .1308, indicating that a player’s pick represents about 13% of the variance in year 4 to 6 PER, which doesn’t seem that high, indicating that there are other factors (luck, good scouting) that can actually be more important than how high a pick is. And, for those of you wondering, the outlier at the top middle of the graph is Giannis.
The following graph shows the same data, eliminating the players with a PER of 0.
This visualization and data give a slightly different picture of the data, eliminating the complete busts. The slope is -0.2, with an intercept of 16.32. As expected, removing the zeroes decreases the slope, as the average PER of the remaining players is proportionally higher for lower picks. While the data and visualization without the zeroes is useful, it does provide a slightly disproportionate sense of the value of later picks, as so many of them are complete busts (as seen by pick 30, where only 5 data points remain).
Both graphs indicate that the first overall pick is worth pretty much a borderline starter, and that the cutoff for that “rotation player” threshold would be about the 10th pick, which has an average PER right above 12 according to the model. After that, we get into players simply on roster (8+ PER), which lasts until about pick 23. From there on in the first round, the average year 4 to 6 PER would be equivalent to someone who won’t last in the league. While this may seem low all across the board, keep in mind how many of these late and even early picks you probably haven’t heard of. It may speak to an overall overestimation of the value of picks by the public. While the first overall pick has a lot of allure, that pick from 2005-2008 was in order, Andrew Bogut, Andrea Bargnani, Greg Oden, and Derrick Rose, only one of which was a true star (and that one being someone who missed multiple years due to injury). And for every first overall LeBron James you also get a LeBron James Jr, or as he prefers to be called, Anthony Bennet.
However, these predictions assume the relationship between pick and PER is linear, which may be inaccurate. Below is the residual plot, which shows how far above or below each individual point was from the linear model.
From the overall shape and disbursement of points in the plot, it seems like a linear model is somewhat appropriate but definitely not ideal in analyzing the value of picks, as the zeroes may throw the model off where it is expecting to go into the negatives. We can also use the linear plot to see which players outperformed their expected pick value by the highest amount. Those players would be Giannis and Kawhi (pick 15), Clint Capela (pick 25), Rudy Gobert (pick 27), and Jimmy Butler (pick 30).
Given that the linear model doesn’t necessarily give the best approximation of picks, we wanted to visualize exactly what level of players each pick was producing. The following visualization takes the tiers we created and identifies how many of each tier each pick had.
The visualization has a lot of interesting insights. The first is the specific prowess of the first pick. The data indicates the first overall pick has about double the chance of being a Hall of Famer as any other pick. Whether that is somewhat of an outlier or that the first overall pick does provide a significant edge over the second pick could be an interesting debate.
Another noticeable trend is that aside from Steph Curry at pick 7, and Giannis and Kawhi at pick 15, all of the Hall of Fame category players are in the first five picks.
While it seems intuitive that the best players would be in early picks, the trend is mostly surprising in regards to picks 6-10. While picks 6-10 are definitely better at producing rotation/starter-caliber players than picks 26-30, the grouping produced only one more All-Star level player than the last five picks of the first round.
To further investigate this trend, we created the same visualization but with each player’s best three seasons. This would allow us to see if it was maybe just specifically years 4-6 where picks 6-10 had similar All-star hit rate as picks 26-30.
This visualization does provide a bit of redemption for picks 6-10. Using the modified years, pick 9 and 10 had significant improvement in All-Star hit rate with guys like Andre Drummond, Kemba Walker, Gordon Hayward, Demar Derozan, and C.J. Mccollum bumping up to the All-Star tier. On the other hand, picks 26-30 did not see the same growth. This could indicate that years 4-6 in specific were just fluky for that 6-10 range, but it is still interesting to keep in mind the analysis from the first visualization.
To further build upon these ideas, we grouped together each group of 5 picks (picks 1-5, 6-10, etc) into a bin with the same tiers.
The results show the drastic difference between a top 5 pick and the rest of the draft. Picks 6-10 have a lower chance of hitting an all-star than picks 1-5 have of hitting a Hall of Famer. However, the starter percentages are similar until the last bin, indicating that the biggest difference is simply the ceiling of these first five picks. Bins 4-5 have very little success overall, and while they have slightly fewer pure busts than big 6, they have overall about the same value. The lack of difference between bins 4 through 6 may indicate teams are simply guessing at this point in the draft, and how high these picks are after pick 15 or so may not be a good indicator of success in the league (making sense of the low R-squared from earlier graphs).
Lastly, we wanted to get a sense of the “typical” pick for each grouping and “sharp” pick. We took the average of the median PER of each pick in a grouping to get a sense of what the typical PER would be in that range and matched that with a similar player. We did the same for the 75th percentile PER instead of median, indicating a sharp or good pick.
The table builds upon our previous analysis, with the “typical” pick being pretty irrelevant after pick 5. For the 75th percentile, you can see that starter-worthy players are available throughout the first round, as “sharp” picks are reasonably good until the last grouping.
Conclusion
Overall, the biggest conclusion to me would be the drop-off after the first five picks. The data indicates that it is quantity, not quality that matters in the early first-round of NBA Drafts. However, it also suggests that quantity may be more important after the first five picks due to the low hit rate throughout.
In terms of practical applications, the data lends itself well to the ideas of Sam Hinkie, and tanking to end up at the top of drafts. While some would argue it has not gone to plan for the Sixers, there is no doubt he had a clear, calculated plan, and based on our research, this plan of ensuring a lack of success to guarantee the top few picks may be better than the Thunder’s current goal of accumulating tons of picks in total. Unless they are able to trade those picks into players, or some end up in the top 5, those picks are likely dart throws. And though the Sixers have not won a championship, they have been pretty much one star away most years, and that can be explained by a low hit rate on their specific picks that include Jahlil Okafor and Markelle Fultz. While it may not make entertaining basketball, and the league does try to diminish tanking with a lottery system for top picks, it does make sense for teams towards the bottom of the league to Trust the Process.
Comments