Jack Kurtz '21
In many NBA scouting circles, foul shooting percentages in college are used to project a prospect’s potential three-point shooting in the NBA. But is foul shooting really a better indicator of three-point success at the next level? A lot of research has been done on this topic, but almost every article focuses only on the players currently in the NBA who have attempted the most threes in a given year. These types of analysis pose two problems. First, they use statistics for only one year of shooting at the NBA level, which has significant variation from year to year, even over an 82-game season. Second, these analyses don’t account for the players who have washed out of the NBA and whose data is equally important, as it allows scouts to establish worst-case scenarios for projecting shots to the next level. One study I encountered in my research, authored by Dashiell Nusbaum, examined the shooting statistics for only those players making at least 2.5 threes per game in 2016. Unsurprisingly, the majority of these players were shooting well above the NBA average, and most were prolific shooters in college. In terms of projecting to the next level, not many of these players had questions about their shooting ability coming out of college. Few scouts doubted that Steph Curry, Kyrie Irving, Klay Thompson, and J.J. Redick would be successful shooters coming out of school. My study aims to take a larger range of players in order to show a more complete picture of three-point shooting projections. This will help to project players like Josh Jackson with significant differences between three-point and foul shooting numbers at Kansas, and whose shooting success could be the difference between a long career and an early washout.
Statistical Methodology
In order to select my data pool, I examined every draft from 2005–2015. I ignored the 2016 and 2017 drafts because the players have not attempted enough shots to stabilize their percentages, for example Jayson Tatum, a below average shooter in college, is shooting 48% from three in a small sample size to begin the season. Then I eliminated any player that jumped from high school to pros or played internationally. Finally, I discounted any player who did not average more than 2.0 three pointers a game for their NBA career. This left me with a sample size of 121 players. Using this sample, I ran regressions for these players using both college three-point shooting and college foul shooting.
The above regression used college 3PT% as an indicator for NBA 3PT%
The above regression used college 3PT% as an indicator for NBA 3PT%
Conclusion
Neither foul shooting nor three-point shooting had an r-squared value above .25; however, it appears that foul shooting is the better indicator of shooting success in the NBA. This could be due to the fact that foul shooting isolates the players’ shooting stroke, removing a lot of variables that come with shooting on the move and with contests from the defense. While foul shooting is a better predictor than three-point shooting, the players who excelled at both were usually near the top of the NBA shooting list. These models use only one factor, so the r-squared values were predictably low, as they fail to account for other variables that affect shooting success such as: percentage of shots that are contested, number of shots off the dribble versus off the catch, and shooting volume. While shooting ability isn’t entirely defined by percentages (just look at James Harden’s 2016 numbers), it is a good starting point for predicting whether a shot will translate to the NBA level. In any case, NBA GMs and scouts use much more data than simply looking at percentages, but this data suggests that they should weight foul shooting success as slightly more important than three-point shooting when evaluating a player.
Additional Analysis and Future Applications
While neither foul shooting percentages or three-point percentages in college proved to be predictive of NBA three-point shooting percentages, I decided to run a multivariable regression using both college percentages. This regression performed slightly better, with an r-squared value of 0.235. However, the fact that combining both percentages still results in a low r-squared demonstrates that college shooting statistics are not predictive without further contextualization. In the future I will attempt to build a more robust model that better predicts NBA shooting success using more than simple shooting percentages.