Predicting Three-Point Shooting for the 2020 NBA Draft Class

Featured Image By Michael Hickey / Getty Images

By Andrew Lawlor

The NBA revolves around shooting these days, so it is crucially important for draft prospects to be able to shoot. We have seen countless talented prospects (Michael Kidd-Gilchrist, Justise Winslow, etc.) falter over the years in the NBA due to a lack of a reliable jump shot.

But it is hard to predict how shooting will translate from college to the NBA. The college three-point line is shorter, for one. For another, the seasons are shorter and amateur teams often play far less open offensive styles, so most players have not taken a high volume of attempts.

One good way to evaluate shooting is to use a statistical technique called Empirical Bayesian probability (see here for a good explanation). Using this technique, we regress a shooter’s three-point percentage towards the NCAA average three-point percentage, weighted by how many attempts the shooter had.

What Empirical Bayesian probability does is regress all players’ three-point percentage towards the NCAA average, but with a stronger regression for players with fewer attempts. The effect is shown in the graph below, with three-point percentage plotted on the x-axis against Empirical Bayesian three-point percentage (lighter dots represent shooters with less attempts):

The basic idea is that, if you have gone 6-for-10 on three-point attempts, you are probably not a 60 percent shooter (so we regress you heavily towards the average), but you probably are better than someone who is a 20-for-100 shooter (we are more confident they are a bad shooter since they have taken a large number of attempts).

A college player’s Empirical Bayesian three-point percentage has a slightly stronger relationship with NBA three-point percentage (0.13 R^2) than his raw college three-point percentage does (0.11 R^2), as shown below:

The leaders in Empirical Bayesian three-point percentage among this year’s college class are as follows:

NameSchool3PA3PEmpirical Bayes 3P percent
Aaron NesmithVanderbilt1156040.2 percent
Saddiq BeyVillanova1757938.9 percent
Desmond BaneTCU1988738.7 percent
Markus HowardMarquette29412138.1 percent
Cassius WinstonMichigan St.1697338.0 percent

Using this statistic, I created a regression model for NBA three-point percentage based on Empirical Bayesian College three-point percentage, Empirical Bayesian College free-throw percentage, Empirical Bayesian shooting percentage on long twos (which includes basically all two point jump shots), Empirical Bayesian College three-point rate (the percentage of field goal attempts that were threes), and position. All statistics were taken from the player’s last season in college. Only players who had taken at least 100 threes in the NBA were included in the training set. I used data from Bart Torvik’s site, which is a great resource for college stats.

The R-squared value for the model is .155, meaning the model explains 15.5 percent of the variance in NBA three-point shooting percentage in the test set. The mean absolute error (MAE) of the model is .028. Since the model is predicting percentages, that means it was off by about 2.8 percent on the average player. The model has a root mean squared error (RMSE) of .039 on the test set. These error numbers are decent, since no one can predict shooting percentage with absolute certainty, but could stand to improve. 

The model is very conservative. It never predicts a player to have an NBA three-point percentage above 40 percent, and only rarely predicts shooters to be below 30 percent. The best prediction in the training dataset came for Tyler Herro, who was predicted to shoot 38.4 percent from three in the NBA (he has slightly surpassed that, shooting 38.9 percent in his career). Luke Babbitt (38.3 percent), Ian Clark (38.2 percent), Luke Kennard (38.0 percent), and Bryn Forbes (38.0 percent) all also had high projections. On the other hand, Andre Drummond (27.4 percent) had the lowest projection in the set.

There have been some misses. Seth Curry (projected 37.2 percent, actual 44.3 percent), Duncan Robinson (projected 37.9 percent, actual 43.7 percent), and Michael Porter Jr. (projected 32.2 percent, actual 42.2 percent) have all shot far better than projected, especially Porter. Porter only played three regular season games in college, shooting 30 percent from three in those games, so his low projection is explainable, and a good reminder that scouting is very important. Relying solely on his statistics did not come close to telling the full story. As for Curry and Robinson, they are two of the best shooters alive, and are outliers that are hard to predict.

Of this year’s college draft prospects, The model predicts Kentucky guard Immanuel Quickley as the best shooter, with a projected NBA three-point field goal percentage of 38.2 percent. Sam Merrill (38.0 percent), Isaiah Joe (37.8 percent), and Markus Howard (37.3 percent) all also profile well. Meanwhile, Precious Achiuwa (29.0 percent), Lamine Diane (30.5 percent), Xavier Tillman (30.8 percent), and Isaac Okoro (30.9 percent) do not project well. 

Among top prospects, Anthony Edwards (33.2 percent) does not project well due to a poor college three-point shooting percentage. James Wiseman (31.3 percent) and Onyeka Okongwu (31.8 percent) also do not project well, but they are big men, so that is more acceptable for them.

Below, see a table of NBA prospects shooting projections, including an 95 percent confidence interval (there is a 95 percent chance they will shoot in that range in the NBA).

NamePredicted NBA 3P percentCBS Sports Draft Ranking
Immanuel Quickley38.2 percent52
Sam Merrill38.0 percent51
Isaiah Joe37.8 percent19
Markus Howard37.3 percent50
Ty-Shon Alexander37.0 percent67
Aaron Nesmith36.7 percent13
Nate Darling36.6 percent90
Cassius Winston36.6 percent39
Malachi Flynn36.3 percent45
Tyrell Terry36.3 percent20
Skylar Mays36.2 percent49
Payton Pritchard35.7 percent47
Tyrese Haliburton35.7 percent3
Elijah Hughes35.7 percent40
Jordan Nwora35.6 percent37
Malik Fitts35.2 percent60
Jalen Harris35.2 percent55
Saddiq Bey34.9 percent16
Grant Riller34.8 percent24
Nico Mannion34.6 percent33
Desmond Bane34.5 percent23
Killian Tillie34.5 percent53
Mason Jones34.4 percent56
Devin Vassell34.3 percent14
Kira Lewis Jr.34.1 percent10
Jaden McDaniels34.0 percent32
Tyrese Maxey33.9 percent11
Patrick Williams33.8 percent9
Jalen Harris33.6 percent55
Tre Jones33.6 percent34
Cole Anthony33.2 percent18
Anthony Edwards33.2 percent5
Obi Toppin33.1 percent7
Cassius Stanley33.1 percent27
Jahmi’us Ramsey33.1 percent36
Devon Dotson33.0 percent35
Zeke Nnaji33.0 percent43
Jalen Smith33.0 percent28
Isaiah Stewart33.0 percent31
Reggie Perry32.8 percent58
Tyler Bey32.8 percent42
Saben Lee32.8 percent59
Paul Reed32.7 percent44
Josh Green32.7 percent25
Ashton Hagans32.5 percent46
Daniel Oturu32.2 percent41
Robert Woodard II32.2 percent29
Onyeka Okongwu31.8 percent6
James Wiseman31.3 percent4
Vernon Carey Jr.31.3 percent38
Isaac Okoro30.9 percent15
Xavier Tillman30.8 percent26
Lamine Diane30.5 percent48
Precious Achiuwa29.0 percent17

A previous version of the above table had a column with incorrect ranges. This column has been removed.

One thought on “Predicting Three-Point Shooting for the 2020 NBA Draft Class

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: