So, I recently came out and said that I'd pick Kawhi Leonard for MVP over Russell Westbrook and his triple doubles. Based on the number of OKC games I've seen, the infamous eye test gives me the impression that Russell Westbrook, when he hunts his stats causes more harm than good to the Thunder. Obviously, a true statistical approach to this question would be difficult and very time-consuming, but I did a rough analysis to gain some insight on Russell Westbrook, his triple doubles, and how they translate to wins.
First bookkeeping note here. I classified each of Russ's games as a Legit Triple Double (LTD), a Barely Got There triple double (BGT), a Nearly Missed triple double (NM), and a Missed It triple double (MI). A BGT is a triple double where the sum of the number of assists and the number of rebounds is less than 22. An NM is a non-triple double where Russ had at least 9 assists or 9 rebounds. Of the 80 games I'm considering (I'm throwing out the last two games of the season because they're pointless), Russ had 28 LTDs, 14 BGTs, 13 NMs, and 25 MIs. That's legitimately insane.
That first pie chart doesn't really do much, but it shows us that the relative sample sizes of these four groups are comparable, albeit small. Now, let's look at how the Thunder fare in these four types of Russ games. OKC generally wins with Russ gets a triple double, but when Russ in on the border, having a BGT or an NMI, his team loses more often than when he has an LTD or a MI. That's a pretty crappy use of the numbers, but it does suggest that when Russ is hunting for that triple double, his team suffers.
How about the margin of victory for these four types of games? It would make sense that Russ is more likely to lose or win close if he's hunting the triple double, i.e. is having a BGT or an NMI.
If Russ is cruising to a triple double or has no chance of getting it, it'll make sense that his team is winning or losing by a wide margin. That trend holds pretty much, as you can see below.
Let's adjust that last bar graph by adjusting for the strength of the opposing team. We basically inflate the margin of victory, so that a loss to a bad team looks worse, a loss to a good team doesn't look as bad, a win against a bad team isn't so remarkable, and a win against a good team is inflated. We basically use the final win percentage of these teams and weight the margins accordingly to produce a rough adjusted margin. Take what you will from this next graph, but it's pretty telling that most of the Thunder's worst losses comes in Russ's NMI games. It's also telling that good wins come when Russ produces a comfortable triple double.
The last quick check I wanted to do was to see how Russ's stats affect OKC's win probability. I did a quick logistic regression and model selection to find which of Russ's stats most significantly affect the win probability. It's all pretty expected, the most obvious of all being that if Russ gets a triple double, OKC is almost 9 times as likely to win. If Russ is engaged defensively and racks up steals, OKC's more likely win. But here's a couple of interesting things from the final model. Russ's scoring output is actually inversely related to OKC's win probability. In fact, every additional point Russ scores after 25 points actually decreases OKC's win probability by about 8%. This rough figure shows that the margin of OKC is more likely to be engaged in a close game, win or lose, if Russ is scoring a lot. That's got to mean something.
This is most telling about Russ's game and its impact on OKC's winning. According to my model, however inexact, for every extra shot attempt after 19 that Russ takes, OKC's win probability decreases by a factor of 90%. On top of that, if Russ shoots under 35%, OKC is approximately 800 times as likely to lose. Granted, this logistic regression hasn't been validated or trained on a lot of data, so some of the results emerge from small sample size theater. But it's just a little more evidence to Russ's extreme usage and stat-padding possibly hindering OKC.
It'll be interesting to see how Russ performs in the playoffs when the pressure of putting up a triple double is gone. Maybe he'll be more likely to make the right basketball play rather than find the assist to complete his stat line. Even so, however critical I've been of Russ, this season has been magical for him. I would love it if the Thunder could poach Blake Griffin from the Clippers and grab a few more shooters this offseason. Russ running amok with a proper pick-and-roll mate and floor spacing could be lethal.
First bookkeeping note here. I classified each of Russ's games as a Legit Triple Double (LTD), a Barely Got There triple double (BGT), a Nearly Missed triple double (NM), and a Missed It triple double (MI). A BGT is a triple double where the sum of the number of assists and the number of rebounds is less than 22. An NM is a non-triple double where Russ had at least 9 assists or 9 rebounds. Of the 80 games I'm considering (I'm throwing out the last two games of the season because they're pointless), Russ had 28 LTDs, 14 BGTs, 13 NMs, and 25 MIs. That's legitimately insane.
That first pie chart doesn't really do much, but it shows us that the relative sample sizes of these four groups are comparable, albeit small. Now, let's look at how the Thunder fare in these four types of Russ games. OKC generally wins with Russ gets a triple double, but when Russ in on the border, having a BGT or an NMI, his team loses more often than when he has an LTD or a MI. That's a pretty crappy use of the numbers, but it does suggest that when Russ is hunting for that triple double, his team suffers.
If Russ is cruising to a triple double or has no chance of getting it, it'll make sense that his team is winning or losing by a wide margin. That trend holds pretty much, as you can see below.
Let's adjust that last bar graph by adjusting for the strength of the opposing team. We basically inflate the margin of victory, so that a loss to a bad team looks worse, a loss to a good team doesn't look as bad, a win against a bad team isn't so remarkable, and a win against a good team is inflated. We basically use the final win percentage of these teams and weight the margins accordingly to produce a rough adjusted margin. Take what you will from this next graph, but it's pretty telling that most of the Thunder's worst losses comes in Russ's NMI games. It's also telling that good wins come when Russ produces a comfortable triple double.
The last quick check I wanted to do was to see how Russ's stats affect OKC's win probability. I did a quick logistic regression and model selection to find which of Russ's stats most significantly affect the win probability. It's all pretty expected, the most obvious of all being that if Russ gets a triple double, OKC is almost 9 times as likely to win. If Russ is engaged defensively and racks up steals, OKC's more likely win. But here's a couple of interesting things from the final model. Russ's scoring output is actually inversely related to OKC's win probability. In fact, every additional point Russ scores after 25 points actually decreases OKC's win probability by about 8%. This rough figure shows that the margin of OKC is more likely to be engaged in a close game, win or lose, if Russ is scoring a lot. That's got to mean something.
It'll be interesting to see how Russ performs in the playoffs when the pressure of putting up a triple double is gone. Maybe he'll be more likely to make the right basketball play rather than find the assist to complete his stat line. Even so, however critical I've been of Russ, this season has been magical for him. I would love it if the Thunder could poach Blake Griffin from the Clippers and grab a few more shooters this offseason. Russ running amok with a proper pick-and-roll mate and floor spacing could be lethal.
Comments
Post a Comment