Strategy Discussions: rotations, play calling, roster management, and more

[bluedevilwildcats]

15 possessions is not just low; it's ludicrously low (~2% of possessions).

Khaman does need to improve his defense but his numbers are being dinged by the players he plays with. No one shares more court time with Kon and Caleb than Khaman. About 53% of Khaman's minutes come with Kon and Caleb on the court. No one else is over 45%. Maliq only shares about 33%.

I agree with CDu that Maliq and Cooper are Duke's best defenders. I would probably put Tyrese ahead of Sion but it's probably coin flip close.

You do understand that insight starts with small sample sizes, correct?

Our most recent starting lineup of Brown, Flagg, James, Kon and Proctor has 55 offensive possessions, that's enough to draw conclusions. And the conclusion is that it's one of the worst lineups because while the defense is stellar, the offense is awful. So perhaps we should try some different, more offensively-minded combos? And guess what, we have, I even listed some of them for you.

The tool I'm using is Evan Miya. Its premium version is used by a ton of college coaches to, among other things, optimize their lineups. Don't know if Jon is using this one specifically, I imagine Duke subscribes to all of the tools.

Sorry for the snark, it's been a long day.

 

[SkyBrickey]

This type of analysis can be misleading. Just one simple point, Gillis and Evans are coming in to play against second teamers or tired first teamers. Who these different roster combinations are playing against matters.

 

[kshepinthehouse]

A couple of observations after this last game:

• There have been a few comments about our offense being a "work in progress." I'm not sure what "work" is being done in practice, but I'm not seeing "progress" in games. If folks are seeing specific offensive changes that the coaches are implementing, can you call those out for the rest of us?
• Several posters have called out that if we aren't in fact a good (38%+) 3pt shooting team, we're in trouble, because we don't have guys who can break down the defense. I think this is accurate, because the defense sagging off our shooters will make it very difficult for us to score.
• Since we can't really "improve" our driving, it seems that increasing our 3P% is critical. To my eyes, that means two of Evans, Kon, Gillis and Proctor should be in the lineup as often as possible. I'd also like to see Harris get a lot of minutes and take a bunch of 3s the next couple of games, as his reputation out of high school was as a 3-level scorer. (couldn't find any high school stats though)

I've taken a look at our lineup combinations with at least 15 offensive possessions. (I know that's a low cutoff). Based on observed efficiency margins, here's what we see:

• Flagg, James, Kon, Maluach, Proctor (overall worst lineup, very poor offense and defense)
• Brown, Flagg, James, Kon, Proctor (recent starting lineup, 6th worst lineup, defense is strong but offense is terrible).
• Brown, Gillis, James, Kon, Proctor (3rd best lineup, very strong offense and defense.)
• Brown, Evans, Flagg, James, Proctor (2nd best lineup, very strong offense and defense.)

These last two lineups have Brown, James, and Proctor in common, so that's a solid starting 3.

Looking at players, pretty much any lineup with Foster doesn't do well, unfortunately. As we know, Maluach has a sizable negative affect on defensive efficiency. While Flagg's efficiency numbers are still strong individually, there are a handful of lineups that perform well without him. I think the key there is to replace him with someone who improves our offense (Gillis, for example) w/o significantly decreasing our defense.

Flagg is really struggling on offense. I think the coaches need to figure out how to get him easy looks close to the basket. I've heard Jon say that multiple times but he hasn't delivered, at least not from what I can see.

So what was our best overall lineup?

 

[jv001]

I don't see any one watching Duke's offense and think that Gillis is a better offensive player than Cooper. Matter of fact, Gillis is not as good as Cooper on defense as well. Gillis is a good, experienced role player. I think the best starting lineup is: Cooper, Kon, Tyrese, Sion and Man-Man/Maliq. I don't know if starting Maliq is better than him come in as a fantastic 6th man. Plus, will this keep him out of foul trouble at the end of games when he's really needed. The 2nd player off the bench is either Evans or Caleb. Neither of them is a point-guard so they need to be paired with one of Tyrese or Sion. What about Mason, well, use him when toughness and defense is needed. Maybe to stop a team on a run. Or pull a Dave Budd to come in and punk an opposing player. Just kidding.

GoDuke!

 

[azzefkram]

So what was our best overall lineup?

It's TBD. By the time you get to Duke's third most frequently used lineup you are already looking at less than 10% of minutes/possessions. By the time you get to fifth (BDW's 3rd best lineup), you are under 5% of minutes/possessions.

A deeper dive into Brown/Gillis/James/Kon/Tyrese lineup shows an Ortg of 167.8 and a Drtg of 55.9. Both of those numbers are great. To get those numbers, Duke had a 78.9% EFG while holding its opponents to 25% EFG. Those numbers are 29% and 35% better than the best and worst rates in all of D1. That doesn't strike me as sustainable.

 

[bluedevilwildcats]

I don't see any one watching Duke's offense and think that Gillis is a better offensive player than Cooper. Matter of fact, Gillis is not as good as Cooper on defense as well. Gillis is a good, experienced role player. I think the best starting lineup is: Cooper, Kon, Tyrese, Sion and Man-Man/Maliq. I don't know if starting Maliq is better than him come in as a fantastic 6th man. Plus, will this keep him out of foul trouble at the end of games when he's really needed. The 2nd player off the bench is either Evans or Caleb. Neither of them is a point-guard so they need to be paired with one of Tyrese or Sion. What about Mason, well, use him when toughness and defense is needed. Maybe to stop a team on a run. Or pull a Dave Budd to come in and punk an opposing player. Just kidding.

GoDuke!

So here is some interesting analysis, to your point about Gillis vs. Cooper.

For each player, how much they contribute to the team is first 3 columns. Next 3 are what their specific numbers point to their impact. And the last 3 columns are what percentage of their impact is based on their specific numbers.

For context: Cooper's total team impact is 7.78. If you compare that to his actual numbers (his PER, etc.), the last column of 58% is what how much of his total impact is explained by his actual numbers. On offense this is 84% (his actual numbers are very close to his overall impact) but on D, only 35% of his impact is from his tracked numbers (steals, blocks, etc.), the rest is explained through lineup synergies, etc.

Compare him to Mason. Only 4% of his Offense, 11% of his defense, and overall 6% of his impact is from his numbers on the court. Basically, 94% of his value to the team is "hidden." I've been pushing for him to get a lot more minutes than his numbers would support, but I was surprised by how much of his impact is hidden.

Looking at it this way, folks with low percentages in the last column (Mason, Maliq, Evans, Patrick) are having a much bigger impact on the team than their tracked numbers suggest. The challenge for the staff is to use this data to test different lineup combinations, even ones that may seem dumb based on visible numbers, to unlock the biggest synergies.

Name	Team Offense	Team Defense	Team Total	Player Offense	Player Defense	Player Total	Visible Offense	Visible Defense	Visible Total
Cooper Flagg	3.65	4.13	7.78	3.06	1.46	4.51	84%	35%	58%
Tyrese Proctor	3.69	2.34	6.03	2.28	0.25	2.53	62%	11%	42%
Mason Gillis	3.78	2.06	5.84	0.14	0.22	0.36	4%	11%	6%
Kon Knueppel	3.17	2.55	5.72	3.11	0.15	3.27	98%	6%	57%
Maliq Brown	2.12	3.49	5.61	0.59	1.16	1.75	28%	33%	31%
Sion James	2.93	2.23	5.16	1.47	0.82	2.29	50%	37%	44%
Isaiah Evans	2.51	2.01	4.52	1.3	-0.22	1.08	52%	-11%	24%
Caleb Foster	2.69	1.2	3.89	1.61	0.29	1.91	60%	24%	49%
Patrick Ngongba II	1.27	2.24	3.5	0.48	0.24	0.72	38%	11%	21%
Khaman Maluach	2.17	0.97	3.15	1.78	0.22	1.99	82%	23%	63%

 

[jv001]

Sounds similar to the +/- stat. Where a player could play badly but be paired with 4 other players that play brilliantly.

GoDuke!

 

[bluedevilwildcats]

Sounds similar to the +/- stat. Where a player could play badly but be paired with 4 other players that play brilliantly.

GoDuke!

Which is why "minutes are earned in practice" doesn't make a ton of sense, as clearly you can't simulate your actual opponents you'll face in each game. But you can use these tools to see which lineups have given your upcoming opponents the most trouble, and then put together your own lineups that closely mimic what has worked best. One of my hopes with Jon was/is that he would take a more data-centric approach than K did. I think he probably has taken greater advantage of these tools than K did / would have, but perhaps not as much as I would like. Still, progress.

 

[scottdude8]

Which is why "minutes are earned in practice" doesn't make a ton of sense (or the corollary, coaches know more than we ever could), as clearly you can't simulate your actual opponents you'll face in each game. But you can use these tools to see which lineups have given your upcoming opponents the most trouble, and then put together your own lineups that closely mimic what has worked best. One of my hopes with Jon was/is that he would take a more data-centric approach than K did. I think he probably has taken greater advantage of these tools than K did / would have, but perhaps not as much as I would like. Still, progress.

I think that the emphasis on "lay-ups and 3s" of this year's team points pretty strongly to the idea that Scheyer is taking a data-driven approach, as that is essentially the current gospel from the analytics.

While I totally understand your defense of the small sample sizes issue, since that's all we have at this point, I do think it remains valid at at minimum needs we need to cautiously approach what data we currently have. After all, all these analytical approaches operate under the law of large numbers. The analogy to me that's always useful here is card counting in Blackjack: yes, if you count cards appropriately, you can shift the odds to be in your favor. But you still need to be able to play long enough for those high-value periods to arise, which is why counting cards if you're just sitting alone playing $5 a hand isn't going to do you much good... instead, you need a huge bank-roll that allows you to play for hours and hours and hours until the odds shift in your direction. A single basketball game doesn't allow you to do that... all you can do is optimize your probabilities, but you're still vulnerable to a relatively small number of random events. IMHO that's why coaching still matters and a blanket approach to the game based solely on analytics won't work in high-level basketball.

I think this is a bit why people may be pushing back on the Cooper vs. Mason comparison. Given his limited minutes in only 8 games, it's much more likely that Mason's numbers are vulnerable to randomness than Cooper's (since he's playing 30+ mpg, more in big games). I think we'll have to look back on this analysis in a month's time to see if these patterns still hold.

 

[azzefkram]

I think that the emphasis on "lay-ups and 3s" of this year's team points pretty strongly to the idea that Scheyer is taking a data-driven approach, as that is essentially the current gospel from the analytics.

While I totally understand your defense of the small sample sizes issue, since that's all we have at this point, I do think it remains valid at at minimum needs we need to cautiously approach what data we currently have. After all, all these analytical approaches operate under the law of large numbers. The analogy to me that's always useful here is card counting in Blackjack: yes, if you count cards appropriately, you can shift the odds to be in your favor. But you still need to be able to play long enough for those high-value periods to arise, which is why counting cards if you're just sitting alone playing $5 a hand isn't going to do you much good... instead, you need a huge bank-roll that allows you to play for hours and hours and hours until the odds shift in your direction. A single basketball game doesn't allow you to do that... all you can do is optimize your probabilities, but you're still vulnerable to a relatively small number of random events. IMHO that's why coaching still matters and a blanket approach to the game based solely on analytics won't work in high-level basketball.

I think this is a bit why people may be pushing back on the Cooper vs. Mason comparison. Given his limited minutes in only 8 games, it's much more likely that Mason's numbers are vulnerable to randomness than Cooper's (since he's playing 30+ mpg, more in big games). I think we'll have to look back on this analysis in a month's time to see if these patterns still hold.

We just need to find out who didn't pass the Rim+3 note to Cooper. My money's on Big Pat as he seems to be the team prankster.

Two more things on Mason. He's the only player that hasn't a negative plus/minus game (you can all "thank me" for willing the incoming -14 against George Mason into existence). Also if you divide his cumulative plus/minus per minutes played it is the highest on the team by a country mile. I couldn't tell you why it's happening, but good things certainly happen when he is on the court.

 

[rsvman]

As the analytics push threes and layups, defenses aim to stop threes and layups.

Therefore, there may come a time, and it might not be too far into the future, in which an offense based on mid-range twos might counterintuitively become the most effective.

 

[Phredd3]

As the analytics push threes and layups, defenses aim to stop threes and layups.

Therefore, there may come a time, and it might not be too far into the future, in which an offense based on mid-range twos might counterintuitively become the most effective.

I believe there has already been a little movement in this direction in the pros. I can't recall the article, but there was one that said several teams are now OK with the right players in the right situations just taking the open twos they are able to get. Certain players make those shots at rates high enough to still be analytics-friendly.

 

[MChambers]

As the analytics push threes and layups, defenses aim to stop threes and layups.

Therefore, there may come a time, and it might not be too far into the future, in which an offense based on mid-range twos might counterintuitively become the most effective.

It's easy to say, but if a team has four or five good shooters on the floor, with some of them good at driving, and the players share the ball, it's hard to stop both three and layups.

 

[Indoor66]

It's easy to say, but if a team has four or five good shooters on the floor, with some of them good at driving, and the players share the ball, it's hard to stop both three and layups.

But there are no teams with that combo, nor will there ever be. Talent is too spread out and performances too inconsistent.

 

[bluedevilwildcats]

I think that the emphasis on "lay-ups and 3s" of this year's team points pretty strongly to the idea that Scheyer is taking a data-driven approach, as that is essentially the current gospel from the analytics.

While I totally understand your defense of the small sample sizes issue, since that's all we have at this point, I do think it remains valid at at minimum needs we need to cautiously approach what data we currently have. After all, all these analytical approaches operate under the law of large numbers. The analogy to me that's always useful here is card counting in Blackjack: yes, if you count cards appropriately, you can shift the odds to be in your favor. But you still need to be able to play long enough for those high-value periods to arise, which is why counting cards if you're just sitting alone playing $5 a hand isn't going to do you much good... instead, you need a huge bank-roll that allows you to play for hours and hours and hours until the odds shift in your direction. A single basketball game doesn't allow you to do that... all you can do is optimize your probabilities, but you're still vulnerable to a relatively small number of random events. IMHO that's why coaching still matters and a blanket approach to the game based solely on analytics won't work in high-level basketball.

I think this is a bit why people may be pushing back on the Cooper vs. Mason comparison. Given his limited minutes in only 8 games, it's much more likely that Mason's numbers are vulnerable to randomness than Cooper's (since he's playing 30+ mpg, more in big games). I think we'll have to look back on this analysis in a month's time to see if these patterns still hold.

Pretty much agree with everything you say here, in a few weeks we'll see a smoothing of the spikes and have a better idea of long-term trends.

There are additional analytics tools. If I can afford them I'll add them to the mix of our analysis in this thread.

 

[SkyBrickey]

One example of how Gillis positively impacted our offense when not scoring in the IW game...

Kon is coming around a Gillis screen on the wing and the defender goes under. Gillis pivots for a second screen to pin the defender down as Kon flares out and catches the pass for an open 3 that he makes. Kon points at Gillis and says, good job. This double screening action is called a flare screen and requires great technique.

My bet is Gillis is a lot better than Cooper at this flare screen action right now to free up Kon, Tyrese, Evans for open shots.

 

[phaedrus]

I believe there has already been a little movement in this direction in the pros. I can't recall the article, but there was one that said several teams are now OK with the right players in the right situations just taking the open twos they are able to get. Certain players make those shots at rates high enough to still be analytics-friendly.

Yes, I think the rule of avoiding the mid-range altogether never really applied to the Kevin Durants of the world. The best shots in basketball are still threes, lay-ups, and free throws, but when the defense has taken those away, it's nice to be able to look for a mid-range shot by your best shooter/scorer.

Translated to Duke, this means Cooper (and maybe Tyrese) are the only ones who should even think about the mid-range, and only where other options have been excluded. Sometimes they are.

 

[Maged]

Banchero was a terrific mid-range shooter. When you are really good at it, you can still make a living off of it. Middleton on the Bucks is one that comes to mind.

 

[SkyBrickey]

Saw some analysis on our shot selection.

As a team, 9% of our shots are mid-range 2s which means 91% are 3s, at the rim, or in the paint. This is a good shot mix.

But 22% of Cooper's shots are mid-range 2s and he's only making 25% of them. These are hurting our eFG% as a team. There's room for improvement here reducing the number of these shots.

Cooper will always lead the team in mid-range 2s because he is our late shot-clock guy, but if Jon can convince him to minimize these by either attacking the rim or passing the ball, our offense will improve.

 

[jv001]

Yes, I think the rule of avoiding the mid-range altogether never really applied to the Kevin Durants of the world. The best shots in basketball are still threes, lay-ups, and free throws, but when the defense has taken those away, it's nice to be able to look for a mid-range shot by your best shooter/scorer.

Translated to Duke, this means Cooper (and maybe Tyrese) are the only ones who should even think about the mid-range, and only where other options have been excluded. Sometimes they are.

I think Kon should be able to take the type shot that Tyrese uses. The guy is extremely talented on the offensive side of the ball.

GoDuke!