Deep Diving WNBA Data – Griner’s Paint Defense

With the decline of Dwight Howard and the dissolution of Roy Hibbert, the top spot in basketball’s rim protection period seems up for grabs. That is, if you’re only focused on the NBA. Expanding our vision a little further, no player has a bigger or more obvious effect on her team’s ability to protect the paint (or more precisely, the restricted area itself) than Phoenix’s Brittney Griner.[1. If you’re That Guy, and it is always a guy, go ahead and x-out and we’ll see you tomorrow. For everyone else, the analysis that follows is of course context dependent, but not much more so than cross-era comparisons of NBA players. By any reasonable measure, Griner’s dominance over her peers far outstrips that of any modern NBA player in this regard.]  This isn’t exactly a novel claim to make about the two-time defending WNBA Defensive Player of the Year, but specific numbers have been hard to come by.

In fact, that difficulty was the genesis for longtime star Sue Bird to pen an article in the Players’ Tribune lamenting the lack of even basic statistical information about the women’s game:

"And so the stories around us, as players, rarely get to become about numbers or stats or performances. Is that because the information isn’t available? Is it because those numbers aren’t logged? If they were logged, would the conversations change? How can we talk and write about the best players if we don’t have raw data that gives context to all of the anecdotal information? What does it mean — what does it really mean — for a little girl to want to play like Dee, or Maya, or even myself?"

Though Bird recognizes there are certain realities at play in terms of the expense required to install and utilize cutting-edge technologies such as SportVU, the issue isn’t about the absence of tracking of hockey assists or miles traveled.  Even a fairly basic question like “who took the most charges in 2015?” is difficult to answer quickly because so much of the data is hidden away.[2. There is a little more information available to the public on WNBA.com, but it is almost a “secret stats” page, not exactly welcoming for new analysts.] The answer, btw was Tulsa’s Plenette Pierson who drew either 25 or 26 offensive fouls.[3. The secret stats site credits her with 26, my own look at the play-by-play data discussed below yields 25.]

However, there is information available for higher level analysis than simply looking at box score scoring categories, though as Ian demonstrated last week even those ground floor-level insights are still useful and largely unexplored. Unfortunately, the data itself presents almost a chicken-and-egg conundrum. Has no analysis occurred because the data is difficult to find and stored in difficult-to-use forms, or do the files remain in that state because nobody has ever taken the time to discover and complain about the problem to the right people? Probably a little from Column A, a little from Column B here, but the point remains basic play-by-play data was not readily available.

This has broad impacts across the board. After all, it wasn’t that long ago (prior to the 2013/14 season in fact) that those performing NBA analysis had to tune and rely upon mostly just play-by-play derived metrics. While not as new and shiny as SportVU and wearable tech, there are vast insights to be gleaned from lineup-level analysis, up to and including regression-based adjusted plus/minus values. My favorite stats![5. No, not really. But this is a case where a good “one-number” metric in the form an Adjusted Plus/Minus or Statistical Plus/Minus model, would represent an leap forward in terms of understanding the WNBA game. To put it another way, this would be foundational research on the WNBA as opposed to the diminishing returns we tend to see when attempting to retrace steps over well-trodden RAPM ground when looking at the NBA side.] And without play-by-play data, and more particularly, PBP data with lineups included this is simply not possible.

In the best of circumstances, creating these files can be a pain. Combined with the other barriers to entry and the relative ease of acquiring similar info for other leagues (NBA or NCAA Men’s) or even sports, it’s not much of a surprise that it largely hasn’t happened before. But, spurred by Bird’s article, it’s time to change that. It took a while[6. As with so much data work in sports and elsewhere, the vast majority of the time is spent on relatively small bits of information. In this case, deriving lineups from the existing play-by-play data wasn’t that challenging. Except for one little quirk. The biggest difference between NBA and WNBA play-by-play data is the former includes a unique player identifier for up to three players on each event. So for example if Al Horford and Andre Drummond competed for a jump ball to open a game and it ended up in Ersan Ilyasova’s hands, we’ll not only have that described as “Jump Ball Horford vs. Drummond: Tip to Ilyasova” but also know that players 201143 (Horford), 203083 (Drummond) and 101141 (Ilyasova) were involved. For WNBA data, there are about 1.5 spaces for such a player identifier – the primary actor (shooter, rebounder, turnover-er, what have you), as well as the player on the receiving end of a personal foul. The other players involved in a play are still there, we just only get on player ID and a lot of text:

“Jump Ball Bone vs Dolson (Thomas gains possession)”

with only the player who won the tap (in this case Kelsey Bone, player ID 203402), specifically identified. Which is more of an annoyance than anything except for one particular circumstance: dealing with substitutions on teams with more than one player bearing the same last name. For example, this past year, the Connecticut Sun employed booth Alyssa and Jasmine Thomas, one of whom, and I have zero way of knowing which, ended up with the ball on this particular play which opened the 2015 WNBA season.

Since constructing lineups requires knowing when players enter and leave the court, not being sure which is which – did you notice the text descriptions only include players’ last names? – is kind of a problem. A problem which affected nearly 1 in 5 games last season, the solution to which took up the bulk of the time parsing out the lineup data. But enough of all that] but that data is now here.[7. Literally here. Download it and work with it, please!]

Which brings us back too Griner, and attempting to capture her impact as a defender. On/Off splits tell us a certain amount. It’s extremely probable she was making substantial contributions in the 2014 campaign when Phoenix’s defense was over 15 points per 100 possessions better with Griner patrolling the paint. That number fell to just over +6/100 in 2015, but the degree to which that was a decline in Griner’s own play, or a knock on effect of Diana Taurasi choosing to focus wholly on overseas play rather than returning to the Mercury for 2015 or something else entirely is hard to tell. With a look into more detailed information, we can look into the evidence of her basket defence.

The principles of seeking to maximize efficiency by getting layups and open threes applies just as easily to the WNBA game as to NBA. The flipside is that preventing and contesting these shots is also at a premium, and this is where Griner dominates.  With some time, I could probably come up with an estimated “points saved” metric, but that would be almost wholly speculative. What we do know is this: the average WNBA team allows about 21.2 rim attempts per 100 possessions[8. In the NBA this number is 27.3 this year, with the difference largely made up in the much higher number of “short midrange” shots taken from just outside the immediate basket area.] and allows 60.8% shooting on those shots.

With Griner off the floor, Phoenix allowed 60.8% shooting on 25.3 rim attempts/100. Even though the percentage isn’t that bad relative to rim attempts, that’s a lot of extra high value, close range shots. Now, with Griner on the floor, Phoenix’s opponents hit only 50.7% on only 18.2 rim attempts per 100. This decline in both efficiency and volume are outliers by a massive degree, and not just in relation to the WNBA but to the NBA as well. Scaling the changes to a percentage of league average for all players with sufficient playing time, produces the following chart:

That seems good, even in a “down” year by raw on/off information.

This is a fairly simple example, but in the coming weeks, we hope to be able to bring more and more in depth analysis from the WNBA game. The good news about nobody doing much of this before is that there are a lot of discoveries to be made.

In addition, we plan on posting more data .csv’s with pbp and other information to allow those of you who are interested to do your own work in this wide open field.