In the first quarter of the Thunder season opener, the announcers were talking about Jeff Green. Hubie Brown I think said “When Jeff Green is playing well, the Thunder are hard to beat.” Obviously, when a player on any team is playing well, we would hope and assume that the team plays better. But how much better? And do some players tend to tip the scales more than others?
I went into this with no good guess about who tends to determine the outcome of Thunder games more than others. You could make the argument that if Durant plays well, we will win, because he’s so much of the offense and when he’s playing well, he’s playing really well. At the same time, if someone like Thabo or Serge lights it up, its more unexpected, so maybe their play is more influential in determining if the Thunder win or lose.
I decided to do this a little more visually and a little less mathematically than I would normally tend to, because it’s a study where its so hard to control for all the variables. So first I went about scoring all of last regular season’s games, to rank which were the most impressive and which were the least. To do this I just took the scoring margin of each game and added or subtracted the opponent’s average scoring margin, to account for degree of difficulty. By this method the November 8th rout of Orlando (by 28!) was our best game of the season, which seems about right, though Vince Carter did not play.
To look at how each rotation player’s performance lead to wins, I put together some heatmaps based on their points, Game Score, and Usage compared to team performance. If you’re unfamiliar with the concept of a heatmap examples are shown below, and for each player and game, the color of the rectangle reflects the intensity of the variable (in this example how many pts scored), with the darker the color the higher the number. The games are ordered from best to worst, so the squares at the top correlate generally with blowouts and wins against good teams and squares at the bottom correlate with bad losses. For my heatmaps I scaled the variables to either the player’s average level or the average level in the each game, they use the same numbers and tell the same story, but scaled to each player’s average is better, because KD’s high numbers take a lot of the contrast away from everyone else.
Above are two heatmaps for points (one scaled each way). In these graphs, because its visual, it’s really easy to draw conclusions, which is nice. However, they aren’t very precise, so there could be multiple explanations for each conclusion, or a conclusion could even be completely wrong. That said, here are some things I notice. Green doesn’t seem to have a high correlation between good wins and scoring. He has some pretty high scoring games in our bad losses. Durant, too, is just all over the place, because he scores so consistently, it seems to not matter that much as a predictor of wins.Our best games, Westbrook doesn’t score many points compared to his average and in our mediocre-to-bad games there is a big patch of dark, where Westbrook scored a lot.
You also notice that Harden and Thabo have dark patches near the top of the graph, it seems like, generally, when we get production from our SG’s instead of Westbrook, things are good. From this graph, though, its hard to tell what is the cause and what is the effect. Does Westbrook look to pass more in those games? Are Harden and Thabo just hitting their shots? It’s hard to tell.
Next, here are the Game Score heatmaps. Game Score is an aggregate metric developed by Hollinger that is a lot like PER, but on a per game level. It takes into account all the standard box score stats and field goal percentage, so it measures a lot of the other contributions to a game in addition to points.
Again, we see Thabo and Harden with a lot of dark bands near the top, showing a strong correlation with good wins. Thabo looks particularly good. In fact, the Thunder were 21-4 in Thabo’s 25 best Game Score games last season. Contrasted to Durant’s 17-8 and Green’s 16-9, that makes Thabo look pretty nice by this metric. But wait! We were 24-1 in Krstic’s best 25 games using Game Score. That’s crazy, and we can see it in the graph. Though it doesn’t look like Krstic’s games lead to blowouts, there’s a solid correlation between color and performance, and it seems we lose a lot of games when Kristic fails to show up (basically a white box)
The usage heatmaps tell much the same story. Westbrook, Collison, and Ibaka generally have a high usage in the Thunder’s worst games. Krstic and Thabo seem to be used more in our wins. Now that seems really strange to me, because it suggests not only that Krstic and Thabo hit more shots in our wins, but they also TAKE more shots in our wins. I’ll leave you with that head-scratcher to think over, because I honestly have no idea how to explain that.
So what does all this mean? Why do some people seem to have a bigger effect on wins than others? It seems like in general it is a bad sign to have a player’s performance correlate strongly with wins. It implies variability in performance, a lack of consistency in areas that the Thunder need all the time. Going one step further with that, it could also imply the player fills unique areas where this is not a lot of depth on OKC.
For example, Harden’s spot up 3pt shot and Thabo’s defense are hard for anyone to replicate and Krstic is one of our only bigs (not counting Green as a big) that can put up big numbers of points. The fact that Thabo, Harden and Kristic seem to “matter” more than others is in fact an indication that these are the positions and roles we need to get more consistent output from, either by improvement of existing players, gameplan, or player acquisition.
PS. Please let me know what you guys think about the ThunderNumbers column, if you have any suggestions on how to improve it, or if you have any statistical curiousness that you’d like to see tackled.
[...] Collard at Truehoop creates something called a Heat Map. [...]