Lots to cover today. I’ll go through the Karl watch quickly and then I want to address a Hoopspeak article several of you passed along and also bring up a fun comment about Karl. Let’s get started shall we?
I think I need to send PopcornMachine.net a Valentines day card. As always, this work is impossible without them.
Now, Karl is without Andre Iguodala and Danilo Gallinari, two of his best players. The bad news is that his response has been to replace them with Corey Brewer and Wilson Chandler, two of his worst players. (Denver Nuggets stats courtesy of The NBA Geek)
One other thing, I’ve been excited (but don’t take too much from that, two games is nothing, we’ll get to that shortly) by Jordan Hamilton and Anthony Randolph. Yes, Hamilton has seen a jump in minutes, but nowhere near enough. After one good game by Randolph, Karl is back to benching him after a few minutes of play. To clarify, I would rather give more minutes to young potentially promising players than players we’ve seen years of terrible play from.
Karl is giving Iguodala’s minutes to Brewer and Gallinari’s minutes to Chandler. Which is maddening, because I say Faried could definitely play more.
As always, Karl is paranoid of playing his second best player the fourth quarter. We’ll talk about that in a second. Long story short. A stupid loss. Yes, a back to back loss against a team with a winning record and with two of your top players is understandable. That said, I won’t be happy with that attitude when our coaching is this terrible.
Faried and McGee?
Jeremy Conlin wrote a piece recently defending George Karl’s rotations or at the very least, defending his hate of McGee. His point is that despite having good numbers, McGee seems to be a defensive liability (at least when paired with Faried). He uses three of my least favorite methods. He uses PER for player evaluation. He uses +/- to examine how good lineups are. He uses on/off lineup data to explain how certain players (e.g. Faried and McGee are bad together)
Now the first issue with the +/- on/off analysis provided heavily in the Hoopspeak article is that it doesn’t control for everything. For instance, who McGee plays with and the players the opponents put out can influence that. And as we’ve discovered in looking at poor metrics like APM, that the number of variables that change and sample sizes we can gather result in data that is essentially meaningless! Now let’s get to the on/off data and Karl’s thinking.
The main thing i notice from Karl watch is that George Karl HATES playing Centers at the end of the 4th quarter. I have a friend at work who is actually family friends with George Karl from way back in his Albany Patroons days, and they stay in touch quite regularly. Being a fan of this site, I recently had him ask him why he won’t play Koufos or McGee more, and he said that he thinks McGee is uncoachable, a boneheaded player and he was against trading for him in the first place. He supposedly got a call from the Wizards coach right after the trade essentially saying “HA! your problem now!”. He said Koufos was “young and inexperienced”. I screamed back ” BUT THEY ARE GOOD!”. haha reason doesn’t always work in these situations.
Here’s the key. Karl has preconceptions about his players. 82games.com has lineup data for all teams. The Nuggets are a few days out of date (as of this post, it was last updated on the 10th) We can see the top 20 lineups by team, how many minutes they played together, and how well they did. Let’s check Faried and McGee
- Lawson-Iguodala-Gallinari-Faried-McGee (2nd) – 98.5 Minutes, Plus Minus of -38
- A.Miller-Iguodala-Brewer-Faried-McGee (4th) – 62.6 Minutes, Plus Minus of +7
- Lawson-A.Miller-Brewer-Faried-McGee (15th) – 33.0 Minutes, Plus Minus of -20
For last year, only one lineup with Faried and McGee cracked the top 20:
- Lawson, Afflalo, Chandler, Faried, McGeee (5th) – 53.9 Minutes, Plus Minus of 0
What we see is a very self-fulfilling prophecy. Karl doesn’t think Faried and McGee can play together and so, they don’t play together. The amount of time they do play together is sparse (the most ‘consistent’ lineup is just over two games worth of time) and as we notice, it’s noisy. Oddly, if I were using the same poor logic to evaluate lineups, it would look like the problem is Ty Lawson, not Faried and McGee. You want a fun “Brian Burke” stat? From January 8th to January 10th, LeBron James had a terrible two game streak. Should we infer much from this? Well, Jeremy does. Here are some of his observations
McGee with Kenneth Faried: -3.4 (338 minutes)
McGee without Faried +2.0 (586 minutes)
McGee with Corey Brewer: +2.1 (660 minutes)
McGee without Corey Brewer: -5.2 (265 minutes)
McGee with Andre Iguodala: +1.9 (565 minutes)
McGee without Andre Iguodala: -3.0 (360 minutes)
Notice above though that that George Karl plays lots of lineups. Only one has over 100 minutes together (Lawson-Iguodala-Gallinari-Faried-Koufos and again Karl hates Koufos in the 4th, wtf?) So that means each of these “with X player, without X player” also includes many other Nuggets and of course, many other opponents. I am sure Karl’s reasoning is similar. Hey, that time I played Faried and McGee with Lawson, Iguodala and Gallinari, it failed! It must be a bad idea. Except, we notice that really none of the lineups have played that much time together. Certainly not enough time for Jeremy or Karl to make such sweeping conclusions (seriously, didn’t we just talk about this?)
And this is a problem I see often in data. We start with a belief, we scan over data to verify our belief and don’t stop to ask how much the data is telling us. I will note that we have some great readers. When we make posts on “short stints”, our fans bring up sample size! That’s huge! Yes, we can analyze and explain what happened. The key is to be careful how much to draw from that. My Karl watches are a great example. I have angrily typed these up game after game. And after many such games, trends emerged (as some astute fans have noticed). And so, when I make my judgement on Karl being a bad coach, it is because we notice that after many games (certainly many more than two) he is making poor decisions.
Is Karl right that Faried and McGee can’t play effective defense together? We don’t know! He hasn’t played consistent enough lineups with Faried and McGee to know either! And when we try to go as granular as Jeremy did, we’re just fooling ourselves. It’s a huge hat tip to both Phil Jackson and Gregg Popovich. When we look at both of these great coaches (who seem to matter according to the data) we notice they avoid overreacting! Contrast this to George Karl, who tinkers with his lineups with the slightest sign of trouble.
I’ve been pretty rough this post, definitely to Karl and now to Jeremy of Hoopspeak. What’s interesting is that I’m actually very impressed by Karl as a person. And I’m sure if I met Jeremy in person (you going to Sloan?) we’d get along well. I’ve noticed that in internet “battles” that attacking someone’s ideas very quickly becomes intertwined with attacking them. And we’ve been trained to believe that attacking people is mean or arrogant. All I can say to this is that I love stats and I love sports. And when I see them used, I want to see them used correctly. Unfortunately, it can be hard to get this right. And candidly, I still make mistakes on a regular basis. I’m lucky enough to have a great crew of fellow writers, and a great group of commenters that smack me into shape when I mess up.
I guess what I want to end with is this. It’s easy to mess up the data. When I call you on it, I’m not attacking you. That said, whether you perceive me as mean or arrogant, it will not change the fact that certain metrics are poor things to use and that certain conclusions drawn from bad data are next to meaningless. So, please George Karl, stop playing such terrible lineups.