Rudy Gobert is Rather Unlikely to Fail

Rudy Gobert had one of the biggest breakout season among NBA sophomores last year. Both his minutes played and effectiveness on a number of different measures rose dramatically. This, after a frankly so-so rookie campaign most notable for his particularly bad turn over numbers. Gobert’s improvement vaulted him to the top of the 2013 draft class, a cohort that has had an underwhelming reputation since well before the first name was called out by David Stern. This reputation has not significantly improved, given the struggles of many of the top picks.

What to make of Gobert’s sudden and relatively unexpected rise remains an enormous question. Can any second year performances from the past better inform us on how to weigh that second year alongside Gobert’s rookie year, age, draft pick and pre-NBA performance?

Analytics people spend an enormous amount of time on draft models, myself included. The draft is the major decision point for NBA teams in acquiring young talent, and the collective bargaining agreement between the league and players’ union gives a number of advantages to teams in terms of holding onto productive draft picks for years to come. Still, player evaluation is an on-going process both for a teams’ own players and players around the league.  Plus, the development of prospects post-draft is a particularly interesting question to me. It is vital for organizations to know which players have real star potential or when to cut their losses with a project which isn’t likely to pan out.

Sophomore Projection Models

To help assess this early career progression, I have developed a Rookie Update model over the last two years to re-evaluate the expected future production and efficiency of NBA players after their first year of play. The nice thing about these models is that they are much more accurate than draft models, explaining 60% of variation in sample with the Rookie Update compared to about 40% with the draft model.  The sophomore models explain even more, approximately 70% of the variation in sample [1. Just below 70% for the efficiency model and just over for the production model]. Of course, it is not surprising that as we get more directly relevant data on how players have performed in the NBA models predicting how they will perform become more accurate!

For all of these models I used the same two target variables; the maximum value of efficiency using a box score metric[1. Alternative Win Score (AWS).] in years three and four of their careers [2. Player’s who fell out of the NBA are estimated at a replacement level] and the max of production weighing minutes played times AWS efficiency similar to total Win Shares[3. Players no longer in the NBA were included in model as having zero production.]. The Production model gives a better separation between stars and efficient roll players, however, the Efficiency model may be more accurate for players such as Shabazz Muhammad, who had their sophomore seasons cut short by injuries not expected to result in lingering harm to their careers.

Both models indicate that the sophomore year has been the stronger indicator for the future than the rookie year. In the case of the Efficiency model sophomore efficiency is 50% more important than rookie efficiency, and minutes played in the sophomore season were a significant factor in the model when pre-NBA production was accounted for [4. The factors were sophomore efficiency, rookie efficiency, pre-NBA efficiency, age and minutes played during the player’s second year in the NBA.]. In the Production model the influence difference between second year production and rookie year was even greater, with sophomore production explaining three times more variation in future production than rookie year when age and draft selection spot are included.[5. Part of the differential is explained by the relationship between draft pick and rookie year playing time, however, sophomore production explains 63% of the model’s predictive power.]

Given both the aging curve and the use of the max score from the third and fourth year there is a general expectation for both efficiency and production to rise within the model,  below are the averages by class for the models6. Using draft classes from 2002 through 2012].

Gobert and the Current Class

This year’s class has mostly caught up to the historic averages after a slightly weaker rookie campaign. [7. Though one interesting side note is the low correlation between the rookie year efficiency and second year in this class compared to previous classes, with a .2 R2 this year from rookie to sophomore year compared to an average of .5 R2 in the prior ten years used in the model, and previous low of .39].

All this points to generally good news for Jazz fans and Gobert, indicating that his breakout season shouldn’t be dismissed. Gobert is projected to be both the most efficient and productive member of the sophomore class moving forward. However, both models indicate that Gobert should expect a modest decrease in efficiency/production. This is due both to his previous relatively unimpressive prior indicators as well as an expected general regression toward the mean from his outstanding 2015 season. Though model projections aside, contextual reasons indicate a significant decline in playing time is unlikely. After all played a only slightly more than  2100 minutes last year while battling for time with Enes Kanter prior to the trade deadline.

After Gobert there is a bit of a fall off with this class.[8. For modeling purposes I have included Nerlens Noel with the rookie class rather than his draft class , consistent with my training model practice] Giannis Antetokuonmpo and Victor Oladipo are projected to be the next highest in terms of production estimates, while Gorgui Dieng and Mason Plumlee making up the second tier in estimated efficiency over the next two years.

Below are the top thirty projected second year players in expected production going forward along with the calculated production this year and the projected change[9. The next thirty sophomores are here, including a fair number who have already dropped out of the league. I have a quick logistic regression model that predicted the players that will fall out of the league after their 2nd year.

]

The signal given by draft selection spot to the model is clear in the table above in the expected increases in production by Oladipo and Anthony Bennett as well as the expected fall by Robert Covington.

The expectations of the Efficiency model are similar, though naturally playing time is de-emphasized. Again Gobert is expected to remain the best player in this class, but the model does not project his efficiency to increase as some other players are.

The takeaway from these analysis for Gobert and other players is how to balance expectations from their last year with sometimes contradictory performance measures in their careers. The second year for any NBA player looks like something of a pivotal year, for both measures in a number of model specifications performance in the 2nd year was the best indicator for the future. However, the other data held insights too; rookie year, age, draft selection and pre-NBA performance all had some weight all indicate we should resist the urge to simply apply an aging curve to last year’s performance.