DATA 621 Blog 3: Predicting NBA Players’ Efficiency
David Quarshie
Intro
In my previous blogs I looked at how the different offensive and defensive stats play a role in a NBA player’s overall efficiency. Offensively, we now know that getting more points won’t necessarily raise your overall efficiency rate but increasing the field goal rates translates to a better efficiency score. Defensively, we saw that the actual stat numbers a player got did more for increasing efficiency. So now that we know which offensive and defensive stats will separately make a player a player more efficient, how do those stats work together? In today’s blog we’re going to see what a NBA player can improve to increase his overall efficiency.
Offense Review
For our offensive model we noted that percentages help improve efficiency over total stats. Looking at the summary of the model we see that we can keep the stats true shooting precentage, field goal precentage, effective field goal percentage, and points.
##
## Call:
## lm(formula = PER ~ ., data = stats_off)
##
## Residuals:
## Min 1Q Median 3Q Max
## -11.1462 -1.7146 -0.2006 1.4715 11.7290
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -8.444752 1.016989 -8.304 8.68e-16 ***
## TS. 45.844088 5.275127 8.691 < 2e-16 ***
## FG. 51.700570 4.268011 12.114 < 2e-16 ***
## X3P. 2.605305 1.213459 2.147 0.0323 *
## X2P. -11.067119 2.433360 -4.548 6.73e-06 ***
## eFG. -44.764416 5.812190 -7.702 6.76e-14 ***
## FT. -1.735443 1.090412 -1.592 0.1121
## PTS 0.004940 0.000263 18.785 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.59 on 523 degrees of freedom
## (64 observations deleted due to missingness)
## Multiple R-squared: 0.7588, Adjusted R-squared: 0.7556
## F-statistic: 235.1 on 7 and 523 DF, p-value: < 2.2e-16Defense Review
For the defensive model we see that the number of offensive and defensive rebounds, along with the number of steals a player gets are very signficant in determining their efficiency.
##
## Call:
## lm(formula = PER ~ ., data = stats_def)
##
## Residuals:
## Min 1Q Median 3Q Max
## -41.808 -2.495 0.068 2.460 18.020
##
## Coefficients: (1 not defined because of singularities)
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.507567 0.590925 11.013 < 2e-16 ***
## ORB. -0.613222 0.384177 -1.596 0.110985
## DRB. -1.264917 0.419477 -3.015 0.002677 **
## TRB. 2.353034 0.789708 2.980 0.003006 **
## STL. 0.416348 0.219731 1.895 0.058610 .
## BLK. 0.358059 0.138297 2.589 0.009863 **
## ORB -0.032064 0.009144 -3.507 0.000488 ***
## DRB 0.024923 0.004078 6.112 1.8e-09 ***
## TRB NA NA NA NA
## STL 0.039041 0.010946 3.567 0.000391 ***
## BLK -0.005164 0.013753 -0.375 0.707427
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.699 on 585 degrees of freedom
## Multiple R-squared: 0.4043, Adjusted R-squared: 0.3952
## F-statistic: 44.12 on 9 and 585 DF, p-value: < 2.2e-16Overall Model
Now that we know which offensive and defensive stats we effect efficiency the most we can make an overall model using them.
##
## Call:
## lm(formula = PER ~ ., data = stats_final)
##
## Residuals:
## Min 1Q Median 3Q Max
## -13.372 -1.821 -0.330 1.553 13.960
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -7.627e+00 7.374e-01 -10.342 < 2e-16 ***
## TS. 3.163e+01 3.635e+00 8.700 < 2e-16 ***
## FG. 3.439e+01 3.700e+00 9.295 < 2e-16 ***
## eFG. -2.805e+01 4.759e+00 -5.893 6.39e-09 ***
## PTS 5.884e-03 5.249e-04 11.211 < 2e-16 ***
## ORB 5.466e-05 5.200e-03 0.011 0.992
## DRB -4.162e-04 2.471e-03 -0.168 0.866
## STL -7.837e-03 6.621e-03 -1.184 0.237
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 3.021 on 585 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 0.7245, Adjusted R-squared: 0.7212
## F-statistic: 219.7 on 7 and 585 DF, p-value: < 2.2e-16Using R’s lm feature we once again get a model for getting efficiency (PER). But this time our results show us that the defensive stats are not that significant in determining PER. However, points, a stat that we knew was significant but had a low intercept, showing that getting more of it would raise PER by much; is now still significant but has an even lower intercept. Other offensive stats like true shooting percentage and field goal percentage are also still signficant but their influence on PER, shown by their intercepts, are extremely lower than when we only looked at offensive stats. For example, true shooting percentage’s intercept decreased 31% when comparing the overall model to the offense only model.
So, when we use the stats we picked to determine PER we see that defense doesn’t really matter. Our model shows us that if you focus on taking and making good shots you’ll be more efficent. It may be terrible for defensive minded players to hear but scoring would most likely make you a better player.