Tracking the Defense

[Nivek]

I'm tracking the defense again, and the results are interesting (as usual). Through the first 3 games, the big defensive problems are pretty predictable -- poor FG defense (team defg of .544), poor defensive rebounding.

The big culprits so far are the bigs -- McGee, Yi, Blatche and Armstrong.

McGee at least is busy -- involved in basically a third of the team's defensive possessions when he's on the floor. He's getting a stop just 45% of the time, though and his drtg is 115.2 (team's is 109.4). The biggest issue with Javale: he hasn't forced misses when he's not blocking the shot. His defg is .532. For the sake of comparison, Haywood's defg back in 04-05 was .328. Even when Haywood's effectivness diminished a bit the past few years, his defg hovered around .430.

Armstrong is reasonably active (23.3% defensive usage rate), and he's better than the other bigs at forcing misses (.484 defg), but he has just 1 forced turnover and 6 defensive rebounds in 40 minutes of court time.

Yi is just awful. His drtg is 133.4. His stop percentage is .376. Seriously, on 32 defensive possessions, the opposition has scored 20 times. His defensive efg is .642.

Blatche is the most effective of the bigs, but he's not an active defender (d-usg rate of 16.4%). His defg is .578, which is preposterous for someone of his size, length and athletic ability. Just ridiculous. His "a little better than the other bigs" drtg is fueled in part by luck -- the opposition is shooting just 3-9 from the FT line off of his fouls. Make that a more realistic 6-9 and his drtg leaps to 119.

Another weak spot is Hinrich. His defg is .598, his drtg is 124.1. He does okay forcing turnovers (1.5 per 40 minutes), but he's generally matching up with bigger players and they're not having much trouble shooting over him.

What's good so far? Thornton is leading the team in defg and defensive rebounds. His stop percentage is .720(!) and his drtg is a team-leading 65.3. His d-usg rate is low (12.7%), which is actually a pretty good sign for a perimeter player (perimeter guys have fewer help responsibilities than bigs). Guys aren't getting open against Thornton, and when they shoot, they're missing.

Wall is generating turnovers at a high rate -- he's forcing 1.6 non-steal turnovers per 40 minutes. He still needs to get better at shot defense, and I'd like to see him help more on the defensive glass, but he's doing okay defensively.

[fishercob]

I love this thread. Kev, can you start us off with a definition of terms and explain your methodology a bit so people can have some more context and get the most out of what you're tracking?

[DaRealHibachi]

Pretty much spot on, your stats go hand in hand with what we see on the floor...

Much appreciated and keep it up!!!

[Wizards2Lottery]

Thornton played really good defense last night.

It seemed like Elton Brand wasn't going to miss anything last night. He made some shots he really wasn't supposed to.

[montestewart]

Thornton has good defensive court awareness, and is the most reliable for help right now, like a roving MLB. He had a fine game yesterday.

[dobrojim]

seems to me (and this is probably obvious to others as well)
that McGee is so confident that he can block shots that he
allows his opp to get open. Someone needs to explain to
him that defense doesn't start when his player gets the
ball.

[Nivek]

The method is basically what Dean Oliver describes in his book, "Basketball On Paper." For those who don't know, Dean is the Bill James of basketball -- The Guru of advanced stats in hoops. Dean wrote BoP, quit his approximately $150k per year job and consulted with the Sonics for a couple years. For the past few seasons he's been Director of Quantitative Analysis for the Nuggets.

In the past, I've tracked a lot of different stuff including play types, court areas, tracking ball movement, deflections, penetrations allowed, steal attempts, and lots more. If you start tracking, even basic stuff, you start seeing all kinds of other stuff that could be tracked. At one point, I was working with a very smart programmer (who's now employed by Portland), who designed a database that enabled me to plow in a HUGE amount of data and pull out some fascinating (and useful) stuff.

In the interest of time, I'm tracking basic stuff -- shots defensed and forced turnovers. Defensive rebounds, steals, blocks, fouls, free throws I get from the box score. Because defense is a TEAM effort, there is splitting of responsibility on various plays. So, if Wall's man dribble drives into the paint, and McGee comes over to challenge the shot, the outcome (whether it's a make or a miss) gets split between Wall and McGee.

If Wall's man travels, Wall gets credit for a forced turnover. If Wall and Thornton trap a ball handler and the ball handler loses the ball out of bounds, each guy gets credit for half a forced turnover.

Basically what I'm doing is building a mirror image of the box score.

I'm probably leaving a million things out, so feel free to ask questions.

As for terminology:

- eFG = effective field goal percentage. eFG is better than FG% because it accounts for the 3pt shot.
- deFG = a player's defensive eFG
- drtg = defensive rating; this points allowed per 100 possessions; the data collected allows me to calculate individual drtg, which mirrors DeanO's ortg/drtg numbers that appear on basketball-reference. (Important to note -- the drtg numbers I'm producing will differ markedly from the stuff on b-r. That's because b-r is using a series of estimates from the box score; I'm actually collecting the data needed to calculate accurate drtg).
- stop percentage is what it sounds like -- out of the player's defensive possessions, how often does he actually stop the opponent.
- Defensive possessions = shots defensed, forced turnovers, fouls that result in free throws, and defensive rebounds. Possessions are shared between teammates -- if Wall defends a missed shot and McGee rebounds it, each guy gets some credit. It's not quite a 50-50 split -- Dean's system divvies up credit based on the team's relative need.
- Defensive usage = percentage of the team's defensive possessions the player is involved in while he's on the floor.
- FTO = forced turnovers. Two types get recorded -- steals, and non-steal forced turnovers. Even though steals are awarded arbitrarily and inaccurately at times, I use the official box score totals. (For example, the steal is usually awarded to the defender who created the loose ball. But, in the first quarter last night, Blatche poked the ball loose and Wall picked it up. Wall got credit for the steal.)

I'm sure I'm leaving stuff out.

[verbal8]

great stuff

[Ruzious]

seems to me (and this is probably obvious to others as well)
that McGee is so confident that he can block shots that he
allows his opp to get open. Someone needs to explain to
him that defense doesn't start when his player gets the
ball.

Google "good shot-blocker doesn't equal good defender", and you get a picture of Javale McGee. Most shot-blockers have a presence that scares opponents from driving the lane. For instance, Wall got by his man a few times in the Atlanta game, but then was tentative putting up his shot, because he saw Josh Smith coming to help out. That doesn't happen against the Wiz. You penetrate, and 9 times out of 10, you get an easy basket. That's a problem with Hinrich defending. He's pesky, but quicker players do get around him - and when they do - he no longer has a Noah back there to help out.

[GoneShammGone]

The method is basically what Dean Oliver describes in his book, "Basketball On Paper." For those who don't know, Dean is the Bill James of basketball -- The Guru of advanced stats in hoops. Dean wrote BoP, quit his approximately $150k per year job and consulted with the Sonics for a couple years. For the past few seasons he's been Director of Quantitative Analysis for the Nuggets.

I find this kind of stuff fascinating. Thanks for taking the time to gather all the raw data. Out of curiosity, how long does it take to track a game? Seems like it must be a big investment of time considering all the detailed action you're looking at. Again, thanks!

Having said that, I have some questions about how to account for team defense effects... for instance you say:

In the interest of time, I'm tracking basic stuff -- shots defensed and forced turnovers. Defensive rebounds, steals, blocks, fouls, free throws I get from the box score. Because defense is a TEAM effort, there is splitting of responsibility on various plays. So, if Wall's man dribble drives into the paint, and McGee comes over to challenge the shot, the outcome (whether it's a make or a miss) gets split between Wall and McGee.

So in this "sharing" situation, do you give both players equal credit or blame for the outcome? Or do you try to make a judgment call as to which player shoulder's the most responsibility? I'm thinking, for example, of a play where Gil's man blows by him to the rim, but then McGee comes over and erases the shot. This results in a empty possession for the opposition, but it doesn't seem right to give Gil and McGee equal credit... is there a way to assign "blame" to Gil, but "credit" to McGee? On the other hand, if the player scores, is it fair to "blame" McGee as much as Gil? After all, Gil was the one who allowed penetration in the first place...

I can see the case for avoiding subjective judgments to keep potential bias out of the process, but I'm wondering if this could lead to misleading conclusions based on some of these team defense dynamics. What's your sense of this? Is this stuff just noise or do you think it can really skew the results you get?

[montestewart]

^
Thanks for taking the time Nivek. Really useful data here for all of us.

[Nivek]

Out of curiosity, how long does it take to track a game? Seems like it must be a big investment of time considering all the detailed action you're looking at.

The more tracking you do, the faster you get. I did the Philly game in less than two hours (started about 10:15 and finished a few minutes before midnight. I make notations directly on a printout of the play-by-play, which speeds things up.

The more detailed stuff takes longer. When I was doing a lot of games, I could get through a game in two and a half hours getting play types, ball movement, court locations and some other stuff too.

So in this "sharing" situation, do you give both players equal credit or blame for the outcome? Or do you try to make a judgment call as to which player shoulder's the most responsibility? I'm thinking, for example, of a play where Gil's man blows by him to the rim, but then McGee comes over and erases the shot. This results in a empty possession for the opposition, but it doesn't seem right to give Gil and McGee equal credit... is there a way to assign "blame" to Gil, but "credit" to McGee? On the other hand, if the player scores, is it fair to "blame" McGee as much as Gil? After all, Gil was the one who allowed penetration in the first place...

I can see the case for avoiding subjective judgments to keep potential bias out of the process, but I'm wondering if this could lead to misleading conclusions based on some of these team defense dynamics. What's your sense of this? Is this stuff just noise or do you think it can really skew the results you get?

In general, I try to avoid subjective calls. It's unavoidable in some cases, but I try to be consistent. To answer your specific questions — responsibility is divided evenly. In the more advanced tracking using the database, I could query the database to see the quality of someone’s help defense. I could look to see how many penetrations a player allowed and the outcome. If the data is gathered consistently, analysis is a lot of fun. As DeanO once told me, the results are robust.

As for the blow-by question, it’s important to record it the same way whether it’s a make or a miss. In most cases like the one you describe, it would be recorded as Gil being the primary defender and McGee as the secondary (or help) defender. I’d note the penetration allowed. While the recording of one particular play may seem unfair at times, that unfairness evens out over time. Plus, then I can evaluate the quality of McGee’s help, for example, vs. Blatche’s vs. Yi’s, vs. Armstrong’s. I can see which guards are allowing penetration and the result. A blow-by directed to the right spot might be less damaging than a contested penetration that violates the force rules.

[Nivek]

I transmogrified the original post into a blog entry.

[Chocolate City Jordanaire]

^
Thanks for taking the time Nivek. Really useful data here for all of us.

+1

[no D in Hibachi]

Simply brilliant work, thanks for sharing.

[TGW]

No one said Mr. Nivek Broom's work was on washingtonpost.com:

A quick look at a basketball box score reveals a glaring deficiency — the info it contains about individual defense sucks. This means that half the game has long been relegated to guesswork, opinion and reputation.

The solution is straightforward: to collect new data. Since the NBA isn’t breaking speed records to track and publish defensive information, I’ve been doing it myself on and off for the past several seasons. The method is based on Dean Oliver’s defensive scoresheet, which he detailed in his book Basketball On Paper.

While in the past I’ve done some geeky tracking (including play types, court locations, tracking ball movement, steal attempts, deflections, and more), this year I’m focusing on the basics — shot defense, fouls that result in free throw attempts, and non-steal forced turnovers (such as travelling, bad passes, ball handling errors, etc.). Steals, blocked shots, defensive rebounds and total fouls come from the box score.

While it’s still too early to draw any conclusions, some patterns are emerging. Unsurprisingly, the team’s defensive troubles have been the inability to make the other team miss (a team Defensive Effective Field Goal percentage (DEFG) of .544 vs. a league average .488) and poor defensive rebounding.

The big culprits so far are the big men – Javale McGee, Yi Jianlian, Andray Blatche and Hilton Armstrong. This is no shock because bigs have the most important defensive jobs, including help responsibilities and protecting the paint. As the bigs go, so goes the defense.

McGee at least is busy, involved in a third of the team's defensive possessions when he's on the floor. He's getting a stop just 45% of the time, though and his defensive rating (points allowed per 100 possessions) is 115 (the team's is 109). McGee’s problem is that hasn't forced misses when he's not blocking the shot. His individual DEFG is .532. Compare that with departed center Brendan Haywood's .328 defg back in 04-05. Even when Haywood's effectiveness diminished a bit the past few years, his DEFG hovered around .430.

Armstrong is reasonably active (involved in 23% of the team’s defensive possessions while he’s on the floor), and he's better than the other bigs at forcing misses (.484 DEFG), but he has just one forced turnover and six defensive rebounds in 40 minutes of court time.

Yi is just awful. His defensive rating is 133. His stop percentage (a measure of how often he prevents the other team from scoring when he’s involved in a defensive possession) is .376. On 32 defensive possessions, the opposition has scored 20 times. His DEFG is .642.

Blatche has not been an active defender (involved in just 16.4% of the team’s defensive possessions while on the floor — a low number for a big). His DEFG is .578, which is preposterous for someone of his size, length and athletic ability. His defensive rating (113) is a little better than the other bigs, but in large part that’s been fueled by luck — the opposition is shooting just 3-9 from the FT line off of his fouls.Make that a more realistic 6-9 and his rating leaps to 119.

The bigs aren’t the only weakness. Kirk Hinrich plays with effort, but has not been effective. His DEFG is .598, and his defensive rating is 124. He does okay forcing turnovers (1.5 non-steal turnovers per 40 minutes), but he's generally matching up with bigger players and they're shooting over him with ease.

What's good so far? Al Thornton leads the team in DEFG and defensive rebounds. His stop percentage is .720 (meaning he prevents the other team from scoring on 72% of his possessions so far) and his defensive rating is a team-leading 65.3. He’s involved in just 12.7% of the team’s defensive possessions while on the floor, which is actually a good sign for a perimeter player (perimeter guys have fewer help responsibilities than bigs). Guys aren't getting open against Thornton, and when they shoot, they're missing.

Prized rookie John Wall is off to an uneven defensive start. He’s generating turnovers at a high rate, forcing 1.6 non-steal turnovers per 40 minutes to go along with his outlandish steals numbers. His shooting defense has been weak, and teams have targeted him. He’ll need to shore up that portion of his game to become a quality defensive player.

http://voices.washingtonpost.com/box-se ... rds_d.html

[miller31time]

I don't know how I feel about having a celebrity on the forum. They tend to lean towards big-headishness. Time to ban Nivek.

(As always, excellent analysis)

[nate33]

Jeebus. The Post will print any hack's stuff these days.

[pineappleheadindc]

Jeebus. The Post will print any hack's stuff these days.

Hi Kev. I *liked* it.

[manifested]

Great stuff. I'm not very good with numbers so I really appreciate when others put the time and effort into this type of analysis. The numbers thus far on Javale really quantifies what people have been saying all along: that when he doesn't block the shot, he has pretty porous defense.

It'll be fascinating to follow this over the course of the season. I'll be really interested to see if Wall's non-turnover defense improves (since he probably won't be getting nine steals on a regular basis) or if we'll be able to find anyone on this team who emerges as an even adequate defender down low.

Weren't a lot of people thinking that Blatche would be a big upgrade defensively over Jamison? I did, but his numbers in this small sample (*coming off injury*) aren't very good. Are his defensive numbers down as well compared to last year, or did I just assume that he was a decent defender because I was comparing him to Jamison who looked consistently physically overmatched in the post?

I was lurking on the board last year but I don't recall what the consensus about Blatche's defense was here.