Checking in on Celtics lineup usage (and other teams)
Re: Checking in on Celtics lineup usage (and other teams)
With respect to the most used lineup, it appears when faced with a top team his tendency is to play it a little or moderately more. An understandable thought given its mildly positive overall performance. But checking the detail, this lineup fails badly in most of these tests and double digit negative per 100 possessions on average. Keep doing it the same way or change?
Re: Checking in on Celtics lineup usage (and other teams)
Bucks:
312 lineups. 73% non-positive.
95% used in 5 games or less. So you get the urge to use a lineup that in 85-100% of the past games you didn't use but you decide this time is the right time? Not guessing well from the results.
63% used in just one game. So many unique solution attempts.
Most used lineup by far is -8 per 100 possessions, worst of any lineup in league used an average of 5 minutes per game for season. Development choice, stubbornness, contract political concerns or ...? Used yesterday for about 15 minutes... and a -10 result.
312 lineups. 73% non-positive.
95% used in 5 games or less. So you get the urge to use a lineup that in 85-100% of the past games you didn't use but you decide this time is the right time? Not guessing well from the results.
63% used in just one game. So many unique solution attempts.
Most used lineup by far is -8 per 100 possessions, worst of any lineup in league used an average of 5 minutes per game for season. Development choice, stubbornness, contract political concerns or ...? Used yesterday for about 15 minutes... and a -10 result.
Re: Checking in on Celtics lineup usage (and other teams)
Bulls
Top 12 selected to be most used lineups over 20 minutes for season, net impact -11. By far most used lineup is slightly negative. Fortunately weak performance at top is offset by impact of dink lineups, at plus 71. Dinking better than average, only 63% of lineups are non-positive. Only 212 were used, lower than average.
Different ways to look weird or ridiculous.
Top 12 selected to be most used lineups over 20 minutes for season, net impact -11. By far most used lineup is slightly negative. Fortunately weak performance at top is offset by impact of dink lineups, at plus 71. Dinking better than average, only 63% of lineups are non-positive. Only 212 were used, lower than average.
Different ways to look weird or ridiculous.
Re: Checking in on Celtics lineup usage (and other teams)
Rockets update
Pretty nice performance by 3 most used lineups. Completely and exactly wiped out for next 5 most used. Top 20 used ends up -64. Dinking better than usual adds only another -6. But neither top nor bottom is where you'd want it. 70% of lineups used are non-positive.
Pretty nice performance by 3 most used lineups. Completely and exactly wiped out for next 5 most used. Top 20 used ends up -64. Dinking better than usual adds only another -6. But neither top nor bottom is where you'd want it. 70% of lineups used are non-positive.
Re: Checking in on Celtics lineup usage (and other teams)
Leaguewide, lineups used over 50 minutes this season are about +3 per 48 minutes. That isn't a lot (and doesn't make a great case for the selection prowess of Coaches) but lineups used less than that are -1.2 on average.
I'd try to use a somewhat tested rotation lineup over as many less tested, dink and super-dink lineups as possible.
I'd try to use a somewhat tested rotation lineup over as many less tested, dink and super-dink lineups as possible.
Re: Checking in on Celtics lineup usage (and other teams)
Celtics update:
49 brand new lineups used in last 6 games. Season is 45ish% over and Stevens is using an average of 8 never before used lineups each recent game. After a brief slowdown, the pace picks up again. Because he just now thought up something better than the first 370 used and / or because after closing in on 2000 season minutes he encountered a totally novel opponent / situation? Searching for better may be a good attitude, but non-positive results for all lineups are holding steady at 67% though, so no overall progress by this data cut of late.
New starting lineup. Old had worked up to a meh (for what you want a starting lineup on a contending team to do) plus 4. Not good enough for long term objectives, so yay for trying something new. But the new lineup is -2 right now. Try it awhile longer- if you really believe it is the most promising alternative of all- then see where you are later this month or in Feb. May have to try one or a few more before hoped for playoffs. 3 of top 5 most used lineups are back to being negative. So it is a choice of turn the new one into a winner, go back to the last one, pick the actual by far best performer in the group (taking Amir Johnson out of starting lineup instead of Sullinger and keeping Olynyk in) or go outside his selected / tried top 5. Going outside the top 5 may be brave and may payoff, but it would mean that the first half of season choices weren't so hot or maybe wise and the re-think is either just in time or maybe kinda late actually. Depends on what you find / develop in next 15-25 games. If you do go outside current top 5, that means trying to discern the best pick among a sea of lineups tested less than 35 minutes, or less than one minute per game. Use hundreds of dink and super dink lineups and not build up more data on a second 5 lineups and you find yourself in the current predicament. Yes, even under a far more concentrated lineup management strategy the minutes of testing will mostly be small, too small, woefully too small but they absolutely did not have to be this small. That was choice or lack of choice. Typical for NBA coaches, but not necessarily acceptable imo. A good lineup simulator might be a plan b workaround but pretty high challenge to beat real testing. (Or maybe not? You control context and maybe some of the subjective attachment to lineup calls.)
They were outside of the playoff ranks last night until the Magic lost and gave them the slimmest edge for 8th. Sole possession of last place on actual wins - expected wins at -3. If they end the season with these things true is Stevens still a great coach? A good coach? A coach with potential to be great or good? Does he have to go 5 seasons with no playoff series wins before the mainstream thought would even question if he is a great NBA coach? Would 5 seasons of negative on actual - expected wins make any impression on anyone else? Still in year 3 so we'll see if it goes that way or not. I am posting data and posing questions, nor making firm conclusions at this point. I am disagreeing with those who have already concluded he is an adequately demonstrated great coach though. His record against the top 10 is a big thing to me. If you can't win a decent amount against them (at his career best this season but still under 30%) your team's season seems unlikely to end in playoff series win(s). That top10 win rate is far below average for teams with 17-21 wins right now and actually below the league median rank for both top 10 and top 16 opponents.
Some may feel my evaluation is too harsh for a 3rd year coach and this team. Perhaps. But lowering the expectations and standards is a pretty big decision. Do you wait til next season or the season after that to apply these standards? Lesser and longer? You can, if you see progress and / or have a lot of patience. But how much progress will this season bring by the end, after adding expensive vets by free agency and trade? I didn't hear a lot of the Celtics' owner and executive talk about expectations. What I did hear played both sides of it. But even doing that, means expectations of improvements weren't, aren't out of bounds. Right now their win% is 4% better than last season. So basically they beat one more team so far compared to last season. Is that enough that the tone of the average coverage (in my awareness) and team self-assessment should be mostly praising, pretty content, maybe even a bit smug given the tiny improvement to date? Are they going to turn a corner? The story continues.
49 brand new lineups used in last 6 games. Season is 45ish% over and Stevens is using an average of 8 never before used lineups each recent game. After a brief slowdown, the pace picks up again. Because he just now thought up something better than the first 370 used and / or because after closing in on 2000 season minutes he encountered a totally novel opponent / situation? Searching for better may be a good attitude, but non-positive results for all lineups are holding steady at 67% though, so no overall progress by this data cut of late.
New starting lineup. Old had worked up to a meh (for what you want a starting lineup on a contending team to do) plus 4. Not good enough for long term objectives, so yay for trying something new. But the new lineup is -2 right now. Try it awhile longer- if you really believe it is the most promising alternative of all- then see where you are later this month or in Feb. May have to try one or a few more before hoped for playoffs. 3 of top 5 most used lineups are back to being negative. So it is a choice of turn the new one into a winner, go back to the last one, pick the actual by far best performer in the group (taking Amir Johnson out of starting lineup instead of Sullinger and keeping Olynyk in) or go outside his selected / tried top 5. Going outside the top 5 may be brave and may payoff, but it would mean that the first half of season choices weren't so hot or maybe wise and the re-think is either just in time or maybe kinda late actually. Depends on what you find / develop in next 15-25 games. If you do go outside current top 5, that means trying to discern the best pick among a sea of lineups tested less than 35 minutes, or less than one minute per game. Use hundreds of dink and super dink lineups and not build up more data on a second 5 lineups and you find yourself in the current predicament. Yes, even under a far more concentrated lineup management strategy the minutes of testing will mostly be small, too small, woefully too small but they absolutely did not have to be this small. That was choice or lack of choice. Typical for NBA coaches, but not necessarily acceptable imo. A good lineup simulator might be a plan b workaround but pretty high challenge to beat real testing. (Or maybe not? You control context and maybe some of the subjective attachment to lineup calls.)
They were outside of the playoff ranks last night until the Magic lost and gave them the slimmest edge for 8th. Sole possession of last place on actual wins - expected wins at -3. If they end the season with these things true is Stevens still a great coach? A good coach? A coach with potential to be great or good? Does he have to go 5 seasons with no playoff series wins before the mainstream thought would even question if he is a great NBA coach? Would 5 seasons of negative on actual - expected wins make any impression on anyone else? Still in year 3 so we'll see if it goes that way or not. I am posting data and posing questions, nor making firm conclusions at this point. I am disagreeing with those who have already concluded he is an adequately demonstrated great coach though. His record against the top 10 is a big thing to me. If you can't win a decent amount against them (at his career best this season but still under 30%) your team's season seems unlikely to end in playoff series win(s). That top10 win rate is far below average for teams with 17-21 wins right now and actually below the league median rank for both top 10 and top 16 opponents.
Some may feel my evaluation is too harsh for a 3rd year coach and this team. Perhaps. But lowering the expectations and standards is a pretty big decision. Do you wait til next season or the season after that to apply these standards? Lesser and longer? You can, if you see progress and / or have a lot of patience. But how much progress will this season bring by the end, after adding expensive vets by free agency and trade? I didn't hear a lot of the Celtics' owner and executive talk about expectations. What I did hear played both sides of it. But even doing that, means expectations of improvements weren't, aren't out of bounds. Right now their win% is 4% better than last season. So basically they beat one more team so far compared to last season. Is that enough that the tone of the average coverage (in my awareness) and team self-assessment should be mostly praising, pretty content, maybe even a bit smug given the tiny improvement to date? Are they going to turn a corner? The story continues.
Re: Checking in on Celtics lineup usage (and other teams)
Do you have any suggestions for fixing Stevens line up issues when accounting for the current roster situation? Do you cut the number of rotation players down to 7-8 and deal with the potential lost trade value of those you bench? Do you believe that there could be a locker room issue if you dramatically cut minutes from 3-4 players in order to consolidate lineups? Do we even care about things like locker room issues/player morale/chemistry? What percentage of the lineups that are negative are are being used for "development"/"trial by fire" for the young players and does this even matter? For this team in it's current construction do you care more about playing the young players in negative lineups or do you play the tried and true lineups at the potential loss of development?
I, like you, worry about Brad Stevens use of these low minute negative lineups, but I want to blame some of that on the roster as it is currently constructed. I have no idea what the team should do (other than stop playing David lee), but I would love to hear your thoughts on this.
I, like you, worry about Brad Stevens use of these low minute negative lineups, but I want to blame some of that on the roster as it is currently constructed. I have no idea what the team should do (other than stop playing David lee), but I would love to hear your thoughts on this.
Re: Checking in on Celtics lineup usage (and other teams)
The first and biggest question is what you are most trying to achieve this year. If it is development, then the rotation should reflect that. I am not going to focus on that initially. Or trying to prop up poor trade value for certain players. Mediocre to bad minutes shouldn't prop up trade value. If the goal is win now, make playoffs and try absolute hardest to win a series, I would do the following:
Rough minutes in non-easy and not getting blown out games-
IT 36, AB 36, JC 36, ET 20, MS 10, DL or JJ 6 (at SF at least until a new wing is brought in), JS 30, AJ 30, KO 30, Jordan Mickey 6. Others only play when winning is considered easy and in hand enough to allow rest / development or you are down by 15 in second half and want to sit guys for not playing well / try something different or just rest main guys. If a big is playing poorly, Lee or Mickey are the first calls. Given they did what they already did to Zeller, he isn't sticking around. But OK for Stevens to use if he now wants to.
I don't have a lot of time to do the lineups fully right now but I'd start with the 8 most used positive lineups and play them as much as possible and then find the best Kelly Olynyk lineups to add to that. I'd pretty much tell Stevens to nearly totally eliminate dink lineups except in case of poor play or injury and even then only when you can't get by with other main lineups. Use around 20 lineups for 85-95% of time and evaluate them with the most minutes of data you can give them, for playoffs and future.
On the trade front, I'd be very open to moving Smart, Turner, Zeller, Lee, Jerebko, Hunter, Young, Rozier. If not dealt for perimeter players and / or draft consideration, I'd let Zeller and Lee definitely walk in summer. Turner I might keep if a handful of big minute lineups with him continue to perform well and the price is right. If not moved for almost anything of value (including cap space / trade exceptions) by the deadline and maybe before if it is not likely I'd waive Rozier and / or Young for a new wing player(s) or a pass first backup PG.
Rough minutes in non-easy and not getting blown out games-
IT 36, AB 36, JC 36, ET 20, MS 10, DL or JJ 6 (at SF at least until a new wing is brought in), JS 30, AJ 30, KO 30, Jordan Mickey 6. Others only play when winning is considered easy and in hand enough to allow rest / development or you are down by 15 in second half and want to sit guys for not playing well / try something different or just rest main guys. If a big is playing poorly, Lee or Mickey are the first calls. Given they did what they already did to Zeller, he isn't sticking around. But OK for Stevens to use if he now wants to.
I don't have a lot of time to do the lineups fully right now but I'd start with the 8 most used positive lineups and play them as much as possible and then find the best Kelly Olynyk lineups to add to that. I'd pretty much tell Stevens to nearly totally eliminate dink lineups except in case of poor play or injury and even then only when you can't get by with other main lineups. Use around 20 lineups for 85-95% of time and evaluate them with the most minutes of data you can give them, for playoffs and future.
On the trade front, I'd be very open to moving Smart, Turner, Zeller, Lee, Jerebko, Hunter, Young, Rozier. If not dealt for perimeter players and / or draft consideration, I'd let Zeller and Lee definitely walk in summer. Turner I might keep if a handful of big minute lineups with him continue to perform well and the price is right. If not moved for almost anything of value (including cap space / trade exceptions) by the deadline and maybe before if it is not likely I'd waive Rozier and / or Young for a new wing player(s) or a pass first backup PG.
Re: Checking in on Celtics lineup usage (and other teams)
Thank you for the quick response. Very interesting read.
I went through the entire thread and I don't think i saw it, but what are you using to generate expected wins? netRTG?
I'm not as versed on these stats when it comes to the NBA, but in other sports expected wins vs actual wins is normally considered the result of chance in close games as opposed to coaching issues. What is your opinion on this idea?
I went through the entire thread and I don't think i saw it, but what are you using to generate expected wins? netRTG?
I'm not as versed on these stats when it comes to the NBA, but in other sports expected wins vs actual wins is normally considered the result of chance in close games as opposed to coaching issues. What is your opinion on this idea?
Re: Checking in on Celtics lineup usage (and other teams)
Here is 42 minutes of a rotation using the minutes I suggested with slight adjustment of minutes for a few players. Product of just a few minutes work. Probably could fill out remaining 6 minutes / do better overall in a few hours, days, weeks.
16 IT AB JC JS KO
4 IT AB JC JS AJ
6 IT MS JC JS AJ
4 IT ET JC AJ KO
6 IT AB JC AJ KO
6 AB ET DL JS KO
All are top 9 most used lineups (regardless of performance) right now, 5 positive to very positive plus the current starting unit at -2. This is not necessarily the best that can be done but it might the quickest and easiest to sell. The main sell is more use rather than new lineup configuration.
One quick way to finish
4 AB MS ET JJ AJ
2 MS RH ET DL AJ
Assuming the last 2 lineups weren't tremendous dogs (compared to current dink lineups) and the top 6 performed anywhere near as good as they have to date, the net pts per game overall would be far better than the current mix with some bad bigger minute lineups and hundreds of mostly negative dink lineups.
The 6 lineups I propose using for 84% of minutes have only actually been used for about 28% of minutes. Even going 1/4th or half the way from current to proposed would be a huge change. Note the small sample sizes and diminish significance of them if you choose but the choice is either try to use that lineup data (with effors to concentrate it for more relevancy) and adjust based on experience OR largely ignore the data and adjust in a non lineup data driven way. Do you really want to stick with option 2? The way that coaches do that leads to about 70% of lineups used on average having non-positive cumulative results? Would a coach who stuck with say 20 lineups (with plus minus results having influence to at least a moderate level) do better than that? I think so. I really doubt it would be worse. I'd like to see. But we may never in NBA. Any relatively high lineup concentration programs in college worth mentioning, studying further?
16 IT AB JC JS KO
4 IT AB JC JS AJ
6 IT MS JC JS AJ
4 IT ET JC AJ KO
6 IT AB JC AJ KO
6 AB ET DL JS KO
All are top 9 most used lineups (regardless of performance) right now, 5 positive to very positive plus the current starting unit at -2. This is not necessarily the best that can be done but it might the quickest and easiest to sell. The main sell is more use rather than new lineup configuration.
One quick way to finish
4 AB MS ET JJ AJ
2 MS RH ET DL AJ
Assuming the last 2 lineups weren't tremendous dogs (compared to current dink lineups) and the top 6 performed anywhere near as good as they have to date, the net pts per game overall would be far better than the current mix with some bad bigger minute lineups and hundreds of mostly negative dink lineups.
The 6 lineups I propose using for 84% of minutes have only actually been used for about 28% of minutes. Even going 1/4th or half the way from current to proposed would be a huge change. Note the small sample sizes and diminish significance of them if you choose but the choice is either try to use that lineup data (with effors to concentrate it for more relevancy) and adjust based on experience OR largely ignore the data and adjust in a non lineup data driven way. Do you really want to stick with option 2? The way that coaches do that leads to about 70% of lineups used on average having non-positive cumulative results? Would a coach who stuck with say 20 lineups (with plus minus results having influence to at least a moderate level) do better than that? I think so. I really doubt it would be worse. I'd like to see. But we may never in NBA. Any relatively high lineup concentration programs in college worth mentioning, studying further?
Re: Checking in on Celtics lineup usage (and other teams)
I used expected wins from Basketball-Reference which is based on net rating.
Actual minus expected wins will be affected by close games, especially against good teams and how much you run up victories, including against lotto teams.
Actual minus expected wins will be affected by close games, especially against good teams and how much you run up victories, including against lotto teams.
Re: Checking in on Celtics lineup usage (and other teams)
Unbeknownst to me Stevens used some of the main 6 lineups I suggested heavier use of for about 26 minutes against Memphis and got very positive results with them. These are his bigger lineups too but not nearly that big usually, my biggest beef / suggestion. Considerably different amounts per lineup and another new starting lineup but the greater concentration on top lineups in this game is encouraging. Much more similarity in the big picture to what I proposed than with his seasonal mix. Will be curious to see future gameflow lineup data.
Still some unexplained thinking, weirdness or randomness to me though. A lineup used 9.4 minutes for season with very poor results suddenly gets 10-11 minutes in this game. And does very badly. Bad enough to amount to double the margin of loss. Ah, tactical button pushing still swings and misses (as usual).
Still some unexplained thinking, weirdness or randomness to me though. A lineup used 9.4 minutes for season with very poor results suddenly gets 10-11 minutes in this game. And does very badly. Bad enough to amount to double the margin of loss. Ah, tactical button pushing still swings and misses (as usual).
Re: Checking in on Celtics lineup usage (and other teams)
Celtics actual wins - expected moves to -4. This level is rare. Even for coaches fired or on the way to that, it usually only -2 or -3. Larry Drew's -5 in 2013-14 is probably the lowest in at least last 10 seasons. He was fired. I found only two other -4s in this period (Scott Brooks in 2012-13, eventually fired, perhaps should have been then and a case with Scott Skiles in 2008-9 that perhaps should have triggered a change then too but he lingered until after he added a few more disappointing seasons). I didn't check everyone, just the most likely.
One game out of playoffs now.
New starting lineup was meh last game. But if BR got updated, no absolutely brand new lineups used. A rare event, maybe even unique so far.
One game out of playoffs now.
New starting lineup was meh last game. But if BR got updated, no absolutely brand new lineups used. A rare event, maybe even unique so far.
Re: Checking in on Celtics lineup usage (and other teams)
Took a look at first 3 seasons for 8 better regarded new generation coaches- Budenholzer, Casey, Joerger, Kidd, Spoelestra, Stotts, Van Gundy and Vogel. Their overall records in first 3 seasons ranged from 40% win to 62% (Casey bottom, Joerger at top). Stevens ties Stotts for next to last. Against Sagarin top 10, the best was Joerger at 52% and worst was Van Gundy at 22%. Stevens takes over the low in this group at 19% so far. Roster quality varies but most had work to do at the start. Trying to compare unequal situations, the data says Van Gundy is the only one of the eight peers with a worse ratio of win% against Sagarin 10 / overall than Stevens. Looking at these 2 win%s, Stevens joins the end of the group (arguably the very end as no one else has 2 finishes in bottom 2), not the middle or top. But we'll see what happens in next few years, for the group and Danny Ainge's "best coach in the country".
Re: Checking in on Celtics lineup usage (and other teams)
Celtics have gone with a second very big minute lineup in last few games. Stevens used 14 lineups last night. The average in 4 randomly selected games earlier in the season was 23. That is almost a 40% reduction. Conceptual progress, I think. But last night the 2nd big, bench lineup wiped out the gains of the starters. For the season they are both just mildly positive. Ok, but are they really the very best choices? Probably not.