Just Shoot More: How to fool Advanced Stats

For this post I’ll be using the Player Comparison tool from Basketball-Reference: 2012 Westbrook vs. Lawson

Lawson vs. Westbrook? Nope, I’m not done. Like an eternal flame, this argument persists. Neil Paine was fashionable late to the discussion but had some very compelling points about why Westbrook is better than I’m giving him credit for. Of course, like all good riddles, when we dig into the data, we find something else entirely. There’s a bit of math to go over. I’ll be hitting Turnover Percentage, then Usage then Assist Percentage (all of these formulas can be found at Basketball-Reference). Enjoy!

Turnover Percentage

Neil makes the following claim.

Lawson is a better shooter than Westbrook, but it’s hard to argue he has better “handles” (objectively, ballhandling being measured by the ability to avoid turnovers). Westbrook and Lawson had identical turnover %’s last season (16.2%) and Westbrook did it with a much higher usage and more touches per minute.

Having a 16.2 TOV% while using 32.7% of possessions is a much more impressive ballhandling accomplishment than having a 16.2 TOV% while using 22.0% of possessions.

While Neil’s numbers don’t match up with those listed at his site Basketball-Reference, this point is true. Lawson had a Turnover % of 14.7%, while Westbrook had a Turnover % of 14.2%. And yes, Lawson’s Usage at 21.2% is much lower than Westbrook’s at 32.7%

A common theme we see is that to fool voters all players need to do is shoot more. More shots means more points. In this case the exact same trick works. Per game Westbrook gets 50% more turnovers than Lawson. So, how is his turnover percentage as low? Well he just shoots more. The equation for Turnover % is:

100 * [Turnovers] / ([Field Goal Attempts] + 0.44 * [Free Throw Attempts] + [Turnovers])

A player can lower this in three ways. First they can get fewer turnovers. They can also get to the line more (which Westbrook does). And, of course, they can just take more shots. Westbrook does this in spades. This ties in nicely with our next them in fact.

Let’s Add in Usage

When the team has the ball, there are one of two things that can happen. They can shoot it (field goal attempt or free throw) or they can lose it (either by steal, losing it out of bounds or a shot clock violation) So we can sum up how every single possession ends by adding these up. And we can figure out which player ended each possession by tracking how many times they turned the ball over or took a shot relative to the team. And this is just what Usage does:

100 * ([Field Goal Attempts] + 0.44 * [Free Throw Attempts] +[Turnovers]) * ([Team Minutes Played] / 5)) / ([Minuted Played] * ([Team Field Goal Attempts] + 0.44 * [Team Free Throw Attempts] + [Team Turnovers]))

Fun equation right? Anyway, Usage is often used as a way to estimate which players handle the ball more. After all, as Neil points out, if I turn the ball over less but I handle the ball less, I may be a worse ball handler.

There’s a few problems though. First, Usage doesn’t take into account assists. Lawson could have a higher Usage rate if instead passing he just took a shot. Second, Usage goes up if you just shoot more. So putting the two problems together. Westbrook turns the ball over a lot. Except his Turnover Percentage is low because he shoots a lot. His Usage is also higher, because he shoots a lot! And guess what? Westbrook is about to show you another way to use the same trick!

Assist Percentage

That, coupled with the fact that their Ast%’s were practically even (30.2% for Lawson, 29.8% for Westbrook), means it’s difficult to objectively argue that “Ty Lawsonhas better ball handling ability than Westbrook”.

Neil’s numbers finally match up with those on his site. Alright, this one was a head-scratcher. Lawson gets 1.1 more assists per game and 1.2 more assists per 36 minutes. How on earth could Westbrook distribute the ball as well? Well let’s examine the Assist Percentage formula:

100 * [Assists] / ((([Minutes Played] / ([Team Minutes Played] / 5)) * [Team Field Goals]) - [Field Goals])

What we are looking for is how often a player assists their team mates shots. Last year the Nuggets made 2572 shots and the Thunder made 2462. Close enough for government work. If Lawson has more assists and their teams have similar totals, how does Westbrook tie him? Let’s look at that another way. Last year, Nuggets that weren’t Ty Lawson made 2198 shots, whereas Thunder players that weren’t Westbrook made 1884 shots.

By taking more shots, Westbrook lowers the total number of shots his team can take and thus ups his Assist percentage! Isn’t math fun?

Summing Up

I’m currently listening to “The Signal and the Noise: Why So Many Predictions Fail – but Some Don’t” by Nate Silver. Nate starts out by focusing on one of my favorite problems – how to use data. A site like Basketball-Reference shows just how much data is out there. This can be overwhelming to some people — like Doug Collins. The important part about is asking what it tells you and using it correctly.

Statistics like Turnover Percentage, Usage and Assist Percentage seem valuable and there is information in them. However, without making sure to examine them closely, we can be fooled in the same way NBA voters and fans are fooled year after year. There is a very simple fact about basketball: missing shots is bad. In doing analysis we can’t ignore this fact. As I’ve shown above, three advanced statistics all go up if a player just shoots more and it doesn’t matter if they shoot well or not. To just take these at face value and not ask: “Is the player actually shooting well?” is a good way to get burned.

To end on a positive note though. I like comments like Neil’s and J.A. Sherman’s (regarding Westbrook’s elite games) When we examine a problem, we have to start with a theory and a criteria for testing the theory. The outcome may not always be pleasant (I am still not swayed that Westbrook had an MVP caliber season last year) but when we ask good questions, we can make sure we get good answers.


Comments are closed.