#analyticsx How to Stop Stephen Curry Wei Gao, Vrushali Walde M.S. in Business Analytics, Oklahoma State University ABSTRACT RESULTS Stephen Curry is a American professional basketball player for the Golden State Warriors of the National Basketball Association (NBA). With 30.1 points per game for the 2015-2016 season, he is the top NBA 3-point shooter and seems to be an unstoppable player in the NBA. He led the Warriors to an astonishing 73 wins with only 9 losses during the 2015-2016 regular season, and a 15-9 record during the post season. The focus of this poster is how to defend him. Some of questions we would like to explore in this poster are: • Can anybody effectively defend Curry? • Is help defense effective when guarding Curry at various ranges? • Will taller and stronger players better defend Curry? • Does Curry have a particular shooting pattern? • Do players with higher salaries make better defenders against Curry? Figure 3 Figure 4 Figure 5 Figure 6 METHODS • Data was prepared and analyzed using SAS® Visual Analytics • The Golden State Warriors played 106 games in the 2015-2016 season. They lost to 10 different teams including the OKC Thunder and the Cleveland Cavaliers. Every single shot by Stephen Curry, in the 44 different games against the teams which they lost to, was recorded one by one through a manual process (840 observations). Variables such as shooting type (jump shot, hook shot, layup, etc.), makes vs misses, shooting range (2pt vs 3pt), shot position from 3pt line (close, one step away, two step away), game result (W/L), etc. were recorded and are listed in Table 1 of Figure 1. Table 2 includes statistics from NBA.com of the defenders guarding Stephen Curry, including defender’s name, height, weight, salary, etc. The data used in this poster were obtained by merging these two tables, containing 17 variables in total. Figure 1 Figure 2 • The FG% in games won was significantly higher than the score in games lost at 95% confidence level (pvalue=0.0124, α=.05). (Fig 3) • The 3P% in games won was significantly higher than that in games lost at 95% confidence level (p-value=0.0368, α=.05). (Fig 5) • About 60% of Curry’s shots are jump shots. 60% of the jump shots are 3pt shots. (Fig 4) • Among jump shots, catch and shoot FG% (50.4%) was significantly higher than non-catch and shoot(41.8%). (Fig 6) • Curry attempted on average 20 field goal shots per game. If the defender could make Curry miss 3.4 more shots per game, the game result would be much different. • Curry attempted 500 3pt shots in these 44 games. He is inclined to shoot 3pt back to back (58.6% of the time) #analyticsx How to Stop Stephen Curry Wei Gao, Vrushali Walde M.S. in Business Analytics, Oklahoma State University METHOD RESULTS Score or not With help defense or not 0 1 Total 0 11.9 35.45 47.35 1 25.4(68.09%) 27.25(43.46%) 52.65 Total 37.3 62.7 100 Does help defense effect differ for 3 point shot and 2 point shot? Statistic DF Value Probability Likelihood Ratio Chi1 1.3893 0.2385 Square Is help defense effective? Statistic Likelihood Ratio ChiSquare DF Value Probability 1 21.8795 <0.0001 Figure 10 Figure 9 CONCLUSION Figure 7 Figure 8 • By the model (Logistic regression, decision tree, neural network), no independent variable was affecting the target variable (make_or_miss) significantly. That means overall, these defenders do not have much effect when defending Curry. No one can absolutely stop Curry from scoring. In that case, research on Curry’s shot and its specific aspects is even more important. (Fig 7) • By the decision tree, when a defender’s weight is less than 182.7lbs, Curry is more likely to take a jump shot. (Fig 8) • Using Cross table, FG% with help defense was 43.46%, without help defense is 68.09%. In Likelihood ratio Chisquared test, variable help_defense was significantly affecting make_or_miss. Thus, help defense was very useful in decreasing Curry’s FG%. (Figure 9) • <help defense> and <shot range> did not have a significant interactive effect on Curry’s FG%. Therefore, no matter if the shot was 2pt or 3pt jump shot, help defense was always beneficial. (Fig 9) • Since the interaction effect of help_defense and shot_type was not significant, help defense aids to any type of shot (jump shot, step away…)(p-value=0.4501). The main effect of shot_type was not significant (p-value=0.5675). (Fig 9) • By general linear model, FG% was significantly different between wide open shots and guarded shots.(p-value<0.01) • By logistic regression, help defense was not effective when Curry takes open shots. (p-value=0.2) • For fadeaway shots, defender’s height was significantly affecting FG%. (p-value<0.001) *FG%:The percentage of field goals that a player makes. • Help defense is very effective on all kinds of shooting at various ranges, except for wide open shots. In that case, do not jump on Curry while he is taking a wide open shot to avoid an additional foul. • Taller players can defend better on Curry’s fade away shot. • Curry is very likely to shoot 3pt continuously. However, for short distance jump shots, or layups, no obvious continuity is observed. So after he scores a 3 point shot, pay more attention for 3pt shot next time he shoots. • Defender’s salary is not significantly affecting Curry’s FG%. Therefore, a superstar salary does not mean a good defender with Curry. • When facing a lightweight defender, Curry would be more likely to take a jump shot. • No one can absolutely stop Curry according to this data. FUTURE WORK AND REFERENCE Future Work: • The data in this report has not included a time variable. If a time variable is added, Curry’s shooting pattern may be found in different quarters of the game. Reference: • The Generalist Bias: Estimating the Value of Three-Point Shooting in the National Basketball Association - Torrey Payne • Predictive Modeling and Statistical Analysis in Sports - Michael Cohen and Matthew Sloane • A Basic Structural Framework of NBA Offense - Patrick Coate
© Copyright 2026 Paperzz