In my deep dive into Basketball-Reference.com, I got tired of looking at player data and decided to dig into the team statistics. It led me to an investigation of the factors that lead to wins. To some extent, I’m just reinventing the Wins Produced wheel, but building it up myself provided me with insights I have a hard time getting from other people’s work.
What I will show are the areas in which the Wolves (and other teams) distinguish themselves. While the resulting recommendations (shoot better from 2-point land, improve defense, reduce turnovers) may be somewhat obvious, I found it interesting to break the Wolves down in this level of detail. I also include a link to a Google Docs spreadsheet that allows you to see a chart for every NBA team breaking down their relative strengths and weaknesses. Details after the jump…
Using data for the last five years (150 team seasons), I estimated a regression model of wins as a function of a number of factors, including:
While I wanted to only include statistics that tied to specific aspects of play, I ended up needing to include DRtg to more fully explain defensive performance. That is, DRB, STL and BLK only explain about a third of DRtg, so separately including DRtg noticeably improves the explanatory power of the model. (It reduces the average absolute error from about 3 games to about 2 games.) In contrast, Offensive Rating (in points per 100 possessions) did not add anything to the more descriptive offensive statistics already included in the model.
Before I get to the statistical model, let’s look at where the Wolves have stood over the last five seasons. The table below shows the Wolves’ NBA ranking for each factor during each of the past five seasons. There are a few areas in which the Wolves stood out in 2010-11, and not always in a good way. They had the fastest pace in the league (which doesn’t end up being a good thing), were third in offensive rebounding, and fifth in 3-point shooting percentage. On the other end of the spectrum, they were 29th in 2-point shooting percentage, 30th in turnovers, 28th in assists, and 27th in defensive rating.
Just these simple rankings tell a lot of the story, but a statistical analysis can help determine the importance of each of these factors in winning games. I did this by estimating an Ordinary Least Squares model with number of wins as the dependent variable and the variables listed above as the explanatory variables. The resulting coefficients are estimates of the change in wins due to a one-unit change in each factor. I calculated an “importance” score for each factor by multiplying the estimated coefficient by the standard deviation of the factor values, or the number of wins a team could expect from improving by a “substantial” amount (one standard deviation) in a particular factor, all else equal. (The “all else equal” part is a little tricky. For example, DRB is a big part of DRtg, but the importance scores don’t reflect that, effectively making DRB look less important than it is.) The figure below shows the importance scores (click on it to make it bigger).
As the figure shows, 2PT%, DRtg, ORB, and TOV are the most important factors. Most of this does not bode well for the Wolves. AST, STL, and BLK are not very important factors in winning.
I next developed a set of charts to demonstrate what sets each team apart from a .500 team. That is, the model predicts that a team with average values for every factor will win 41 games. Deviations from the average value contribute to a team being better or worse than average. The figure below shows the resulting figure for the Wolves. (The vertical axis is the number of wins gained or lost by being different from the average value over the last five years.)
Not surprisingly, the Wolves are hurt most by three factors: low 2PT%, high TOV, and high DRtg. That is, they miss a lot of shots inside the arc, they turn the ball over a lot, and they don't play much defense. Note that the high pace of play has cost them about 2 wins. They actually pick up some wins by shooting well from 3-point land and gathering a lot of offensive rebounds.
There is an additional "factor" on the far right of the figure called "unexplained." This is the difference between the actual number of wins and the number of wins predicted by the model. The model predicts that the Wolves would have won 22.2 games with their statistical profile. This is consistent with Pythagorean Wins (23.9), Wins Produced (22.8), and Win Shares (24.9). In other words, the Wolves were bad in a way that a variety of statistical models can't explain. It could be due to bad luck, bad late-game coaching, late-game boneheaded play, etc.
I've made one of these for every team using 2010-11 statistics. It can be found in this Google Docs spreadsheet. (I had to re-format the figure to conform to the limitations of Google Docs, which are very annoying).
In the spreadsheet, use the highlighted cell as a drop-down list to select the team. Here's a short summary of some interesting (mostly obvious) things I found:
I find it interesting to compare across teams, as the figures provide a handy snapshot of what they do well and poorly. I hope you have some fun with it as well.