dougthonus wrote:MikeDC wrote:So here's my defensive rankings, based on positional averages for guys who played at least 500 minutes last season.
All of these ratings are based on standard deviations from the positional average. So if a guy ranks +1 in a stat, it means he's 1 standard deviation above the average. If he's -1, he's a standard deviation below the average. If he's 0, he's right at the league averages for his position.
I evaluated Ayo, Huerter, Terry, and Okoro as G-Fs.
Defensive Usage (overall how many shots did the player defend per 36 mins). More is better. Guys who can and do soak up lots of defensive possessions and defend well are what it means to be a good defender.
- Heurter 0.99 (defended 15.29 shots per 36)
- Terry 0.5
- Okoro -(0.22)
- Dosunmu -(0.79) (defended 12.47 shots per 36)
Huerter is the best here, with Terry second.
You can further break this down by shot type, 3 pointers, and 2 pointers and you can look at the DFG% and contest rates
3 Pt Usage/DFG%/Contest Rates
- Heurter 0.24 / 0.22 / (-0.1)
- Dosunmu (-0.72) / (-0.45) / 0.33
- Terry 0.80 / 1.17 / 0.78
- Okoro 1.13 / (-0.06) / 0.22
Terry is the best here, and the only one that's clearly above average across the board. Okoro is second because he's average on higher usage. Huerter isn't high usage, but he's average, not bad across the board. Dosonmu is the least effective at defending the 3.
2PT Usage / DFG% / Contests - Heurter 1.43 / 0.75 / (-0.04)
- Terry 0.18 / 0.9 / 1.29
- Okoro (-1.2) / 1.08 / 1.48
- Dosunmu (-0.42) / (-1.18) / 0.36
Huerter, Terry, and Okoro all got good results here. Okoro got the best results, but on the lowest volume. Overall, Terry gets the slight edge, but maybe guys just don't even try to drive against Okoro? Most likely, Okoro is benefiting from having Allen and Mobley behind him. The Cavs actually wanted to funnel guys into those guys swatting away shots.
Finally, possessions is a huge part of defense. A really simple measure is (Loose balls recovered + Deflections + Drawn Charges + Steals + Blocks ) / Personal Fouls. If you look at the top guys in the league for this, you will see pretty much the best defenders with a couple of exceptions of slow guys who literally just defend by taking charges (looking at Mo Wagner here).
- Heurter 0.45 (basically slightly above average by not being bad at anything.
- Terry (-0.41) (good in steals, deflections and loose balls, but really fouls a ton)
- Okoro (-1.26) really doesn't generate many takeaways at all. He's a good man defender, but really below average here and notably worse than all the other guys. Again, kind of weird.
- Dosunmu (-0.53) Rates out similar to Terry, but where Terry is very active but foul prone, Ayo is basically just below average at most things and not foul prone.
This is just one season, but all of these guys played enough minutes that it's' a pretty representative sample of what they can do. Generally speaking Ayo is consistently the last guy or near the last guy I'd pick to play defense.
I'd probably pick Terry first over Okoro to guard the point of attack.
All things considered, Huerter, though, not far behind and in many ways a more solid defender, and he's light years better offensively.
This requires a lot of faith in DFG% being meaningful.
No it doesn't. DFG% it's not getting any more weight that the other factors. It's a holistic evaluation based on 4 different factors that obviously matter.
Huerter has the lowest contest rate of the guys and visually looks like the worst defender.
And yet, his contest rates are within a couple hundreds of a standard deviation from the positional average. This is a common problem with statistical interpretation. People want to say what's best vs worse, but if you've got 3 guys with contest rates (on 3 pointers) within a half a standard deviation from the average, there's not much functional difference between them. Ayo contested 45% of 3s, Okoro 44% and Huerter 41%. The league average for the position is 41% with a 9% standard deviation. It's a fairly trivial difference.
I'd guess his higher rate of shots defended is based on defenses targeting him and his low contest rate not because he's positively soaking up defensive possessions.
Except his contest rate isn't actually "low". It's an average contest rate. The proper frame of reference isn't the other 3 guys, but the entire league and, in the context of being targeted, the other guys on the team. Think about this from the perspective of a team playing the Bulls. Do you target Huerter or Giddey, Vuc, or Coby? Huerter is the 4th guy you'd want to attack if you're the offense.
The Bulls game is the opposite. The defense wants to funnel defensive possessions to their best defenders. It's crazy that in Huerter, who is average, is one of their better defenders, but he is. Part of the reason the Bulls were better than they should be defensively is that, relatively speaking, they were pretty good at this last year. if you look at total shots defended/36 (which are a little inflated in general for the Bulls because they played fast) the guy that really got targeted was Giddey. While they did a pretty good job of "hiding" Coby (and Ayo and Pat, who was terrible). Matas, Collins and Huerter were "Plus" defenders in that sense.
I don't know if I fully understand the methodology you are employing here, and I haven't done enough robust looking at DFG% (which I'd imagine is incredibly noisy for a lot of reasons), but I think the data you posted is likely subject to be interpreted in a lot of ways.
People generally interpret things how they've determined they want to interpret it.
From your possessions calculation, it's worth noting that a drawn charge and a steal is a possession generating event, a loose ball, deflection, and block are not, they require other events to happen to also generate a possession.
Simplicity. I've got a weighted version that accounts for this but it's a hobby, not a job.
Why would you divided by fouls drawn rather than subtract fouls drawn? A foul drawn has no reason to be a divisor, and a foul drawn is also not always a possession lost. A lot of time you're simply resetting the clock on a possession, and you'd also need to subtract out intentional fouls, or fouls you save points because they make sense to foul a guy whom otherwise is going to dunk and it may save you points. A foul is not always a negative event.
There are a couple of technical reasons for treating fouls as negative:
1) The vast majority of fouls are negative events, and they are most directly related to the ability of a defender to do the other, good defensive plays. There's an inherent rate between committing fouls while trying to do good things.
2) The "good fouls" are generally stochastic (random), especially if you are analyzing things by position, which I am. If everyone is doing these things in the same proportion, they can be ignored because they add (or subtract) no useful information. That is, you don't actually need to subtract something out if it doesn't change the bottom line. Good fouls are roughly equally distributed, so it doesn't matter if you leave them in or take them out.
3) In addition to the intuitive point of 1) fouls are better as a divisor because, since there is noise in fouls, Good things/bad things (fouls) creates a lower variance than Good things - bad things (fouls). The standard distribution of Good/Fouls is about 33% around the mean values. The standard distribution of Good-Fouls is about 63% around the mean values. So it does a better job of losing the noise that does exist.
In reality though, it really doesn't matter too much... you could subtract or divide and the answer would still be a meaningful, informative result...
The possession stats themselves are also reasonably noisy, and while I completely agree valuing possessions is important, without really drilling down into what creates a true possession accurately, the outcome is garbage in / garbage out.
Not at all. You don't need to create a false sense of precision when a good approximation gets you most of the way there. Like, here are the top guys with divide and here are the top guys with subtracting fouls
Subtract / Divide
Wemby (1) / (2)
Daniels (2) / (4)
Caruso (3) / (23)
Isaac (4) / (16)
Ausar (5) / (44)
============
Butler (9) / (1)
Wright (15) / (3)
Kawahi (10) / (5)
Just eyeballing it, maybe subtract is better, but it might just be confirming visual biases. The fact that dividing reduces the standard deviation so much suggests that's the better method for a simple analysis, but in the full version where I'm weighting everything, I'd be subtracting anyway.
But getting fixated on which one is better is getting lost of in the weeds when the overall view that both methods present is pretty clear.
It's the same reason a ton of stats massively overvalue defensive rebounding on the theory it is worth a possession when you should probably be looking at marginal rebounding differential since 75% of rebounds are defensive and the defense is generally expected to get the rebound, so it isn't worth a full possession.
Agree. As I said, this is a simplified version. But 10% of the effort generally gets you 90% of the answers, and while it's reasonable to further refine it, it's not going to dramatically change the results.
I don't mean to be critical, because I do find everything you posted really fascinating, and it's hard to work up novel approaches to measuring defense, and I appreciate your effort, but in the end watching basketball for 5 minutes, Huerter is clearly the worst defender of this group.
I think this sums up why reliance on "the eye test" is ridiculous folly. The hubris of thinking you can watch 5 minutes of ball and make a definitive pronouncement like this (even jokingly) is off the charts.
But the real issue is that the eye test puts you at odds with reality and the analysis I've done explains reality. Here's reality:
1. The Bulls were demonstrably better defensively with Huerter than, say, Ayo. The stats I've developed inform us why, and the eye test doesn't.
2. The Bulls hall of fame coach known for his defense was perfectly fine with Huerter and in fact preferred him to other options. Likewise, he didn't have much use for Jalen Smith despite the occasional flashy plays that the eye test seems to overrate.
3. Being the "best" defender doesn't always mean something consistent. Do you mean playing the point of attack? Defending the rim? Being adequate at everything? By breaking things down, we can see why some things matter more than others, and to what degree. And how stuff fits together.
4. Okoro generates very few takeaways and disruptive defensive plays. He's a good man defender, but on a way below average number of possessions. Like with his shooting, he's coming from a team that created a lot of positive situations for him to a team that is going to rely on him to be a game changer. My one prediction is that this isn't going to work out very well. But again, my analysis indicates obvious reasons for this. While the eye test is just "look, that guy defends hard". It's not unlike with Jevon Carter a couple years ago. Everyone looks really good when they've got Brook Lopez and Giannis behind them. Less so with Vuc and DeMar.
in every case, diving into and including all the numbers provides a reasonable explanation for WHY we are seeing the decisions we see.
Watching for 5 minutes, or even if you watch for 48 minutes x 82 games, is pointlessly excluding a huge amount of information because the human mind literally can't store and classify that much data. People remember anecdotes and highlights. But every player in the NBA, and every prospect probably, has a set of highlight reels full of anecdotes and highlights. So what happens is, if you just rely on that, you end up confirming your biases. And, it's really hard to watch a game and understand good defense. You can see that here, where about 75% of the time someone calls out a bad defensive play, someone else will jump in and argue it was really someone else's fault.
So there are two really big problems with "watching"
1. correctly interpreting what you see.
2. correctly remembering and evaluating all you see
People suck at both of these things.
That's not to say one shouldn't watch, but in light of these known limitations, the better approach is to rely on the bigger, objective dataset and then look at how what you see fits into that reality. Relying on what you see first, and then arbitrarily throwing out the bigger, more objective sample of data is just throwing science and reason out the window in favor of comfort.
I'd say undervaluing contests
Contests are like everything else, not unambiguous. A guy who contests literally everything is like a guy who tries to block every shot or always gamble for a steal. A flaw of watching is that people systematically overvalue the obvious.
You'd also need to factor in quality of opponent defended into the mix here to really drill down into things.
Not really, everyone is being measured against league and positional averages. At some point it might matter, but it's really far down the list.
