APBRmetrics

Rockets just plus 38 with 5 best lineups and minus 130 with rest of lineups. So they need major improvement at the top and with rest. Only one lineup used more than 3 minutes per game and none used in half of games. Injury game loss is moderate but come on. Looks like no plan or constantly changing ones. 6 of top 14 lineups on minutes are negative. Fix or drop them. Just 32% of lineups are positive cumulatively so the team is not playing well and lineup selection for situation is not very good. Warriors and Spurs play their 5 best lineups more than 50% more often. No guarantees good early results will continue but it may be one of the better ways to guess, on average.

Looking at McHale / Bickerstaff splits would be useful. It appears Bickerstaff has almost completely stopped using McHale's two best performing lineups, each including Lawson. Bickerstaff's top 4 minute lineups? 3 negative, one positive. 6 of top 10 negative. Overall still only 1/3rd of lineup choices are positive. 2 lineups used an average of 4.5 minutes per game each. A baby step toward greater concentration but big steps still needed. Better win-loss but how much of that is schedule and luck vs. improvement and specifically improvement in lineup management? Would need more and closer study.

Celtics lineup analysis update:

Most used lineups are performing better.

But starting lineup net pts per 100 possessions is just 32% of the cumulative RPM estimates and only the 102nd best performing Celtics lineup, though this is affected by quality of opponents faced. Only one lineup used over 10 minutes for season is in top 50 performing Celtics lineups. Random very short minute lineups will unsurprisingly occupy a lot of these slots but just one lineup over 10 minutes or used more than 24 seconds per game in top 50 performers?

% of non-positive lineups is at 67%, up from the 62% almost a month ago.

99.66% of the 301 lineups have been used in less than 1/3rd of games. Only one in over half. Only 2 lineups used in much over 1 minute per game. 91% of lineups used less than 10 minutes total, 82% less than 5. Celtics have used 30% more lineups than league average. # of brand new, never before used lineups has slowed considerably but they are still getting introduced.

Plenty of teams doing worse, but there appears to be room for improvement here, I think.

Top 20 most used lineups are plus 113 cumulatively. The other 281 are -10.

Actual - expected wins still at -2.

There is so little known or at least published on lineup performance and norms.

Checking GSW and SAS, their biggest minute lineups were 77th and 109th best performers of all lineups for their teams so the Celtics performance here is inline. GSW had a whopping 3 lineups over 10 minutes use in top 50 performers. SAS zero.

It might be better to look at biggest minute lineup performance vs. just the bigger minute alternatives. Among the 25 Celtics lineups used over 10 minutes, the biggest minute lineup was 14th best performing. Over 20 minutes use, 6th of 10. Ok maybe, but not that impressive. SAS and GSW's most used lineup was only 14th and 15th best of 25 lineups over 10 minutes, so I guess even the best aren't particular impressive on this measure. Even this level of sample size is too small for a meaningful comparison. But searching for, checking and evaluating potential measures? I think that could be worthwhile. Opponent strength adjusted lineup data would be worthwhile addition.

What are the chances that a top 5-10 most used lineup would perform better than the most used lineup for a team in equal minutes and same circumstances? I dunno, but it may be pretty good in many cases. There is so little in the way of meaningful testing to try to find out, in large part because of all the dink lineup use with demonstrated poor returns on average. I am for reality testing more lineups over 200-500 minutes first and foremost but I'd also want to test 100 to 1000 lineups against 20-100 lineup types for 1000 plus minutes of simulation each. And then after the first 2-100 million minutes of simulation and hundreds or thousands of hours of evaluation, do more, constantly trying to improve the quality of the simulator and the simulator data & its guidance. And the degree of coach consideration and utilization of the guidance, provided positive impact is experienced.

Is you gave Head Coaches and GMs a quiz on how their 5-10 most used lineups did overall and against 5-10 opponent lineup types to the level of positive / negative or positive/ near neutral / negative for a full season how well would they do? Maybe not as well as they think they would do. Has any team tried it? Probably not, I'd guess; but maybe I'm wrong about the accuracy of awareness. Randomness in the relatively low minutes would hurt predictably but I think it would be worth seeing how poor the predictions were. Maybe there would be more support generated for more testing, real and simulated. Even if lineup matchup performance is difficult or improbable it is probably worth realizing that ignorance / uncertainty deeper and trying harder to get at least a little bit more accurate and optimizing. If even the best Coaches and analysts are limited in predicting the performance of the biggest minute lineups in a handful or two contexts, perhaps it would be better understood how ridiculous it is to claim to be able to use dink lineups reliably in specific situations as it appears Coaches implicitly claim to be able to do by trying it so much.

Is there more application of Bayesian techniques to lineup analysis to be done (beyond what JE and Joe Sill have done previously)? Daniel Myers mentioned possibilities years ago. There are a few papers out there that probably haven't been discussed enough or used as inspiration enough. https://www.stat.sfu.ca/content/dam/sfu ... u-2007.pdf
http://www.cs.toronto.edu/~gdahl/papers/dpmfNBA.pdf
http://blog.smellthedata.com/2011/02/th ... l.html?m=1
http://www.apbr.org/metrics/viewtopic.php?f=2&t=26 (at team level)
http://espn.go.com/blog/playbook/dollar ... /undefined

Pehaps I am forgetting or unaware of other relevant previous discussion.

Does it matter a lot to apply Bayesian techniques to lineups instead of players? How? Why? How much?

Anyone hear this presentation or have access to it?
https://www.amstat.org/meetings/jsm/201 ... tid=312419

Is going beyond simple logistics classifer worth it? http://arrow.dit.ie/cgi/viewcontent.cgi ... 0%26sa%3DN


This is probably an important start out there. http://journals.plos.org/plosone/articl ... ne.0136393 Player tracking data, skills plays... Newly published there; not sure if it was shared previously.

There is more out there from James Piette and others but I'll stop here (at least for now).

Celtics actual wins - expected is current tied for league's worst (with Thunder and 76ers) at -3.
Stomping lottery strength level teams at a good W-L clip and running up the score more than usual. Only 25% win against Sagarin top 10 though, for just 20th place on that measure.

Last 3 games, 19 more new never before used lineups. Was there really a good reason for all of these at this point?

Fully a 1/3rd more lineups used so far than last season by end of Dec. with a few more days to increase the margin.

Despite all the time given dink lineups the Celtics have the 7th most used lineup in league, albeit at a bit less than 8 minutes per game for full season. Just barely an above median and mean performer amongst the top 10 and top 25 most used lineups though. Not getting as much edge from this lineup because of time and quality as you'd like to get and as the best teams generally get. The most used Celtics lineup makes only the 19th biggest net point contribution to winning (plus 15 for season, about 2/3rds of a pt per game- when used) amongst big minute lineups (over 100 min.) in league and only 91st biggest amongst all lineups. That is not enough to get where they want to get.

I don't know what kind of reception Brett Brown is getting from local fans but the national media, imo, has been quite soft on him. Probably too much sympathy for a tough situation and being a Popovich lieutenant. But he has the worst win% (19.5% and steadily dropping) of any coach of more than 82 games in the history of the league. Could you really do worse or much worse than that? The early extension of him appears to me a continuation of the tank or unwise / undeserved. I don't hold him blameless for out of control young players either.

Just two lineups used more than a minute per game, only one used in more than the half the games. All of the top 6 chosen to be most used lineups are worse than -13 per 100 possessions. 380 lineups used (so at least one coach has used more than Stevens), 74% non-positive cumulative results. Only 9% of lineups given more than 10 minutes run for season. Could lineup management look much weaker and lacking in insight if it was completely random? Is this any way to learn and build to a brighter future?

Billy Donovan appears to be doing well with his most used lineups. At least on average- 9 of top 10 most used are positive and the other is barely negative. But weak record against top 10 and just average against top 16 may suggest there still are issues in those important subsets. 66% of his 212 lineups are non-positive for season. Sure looks like concentration / more concentration is the prima facie, big picture right choice for way to go on lineup management here, too.

Popovich. Considered by most the reigning master and one of top 4 of all-time.

316 lineups used. Just 3 for more than 2 minutes per game. Just 3 for more than half the games. 66% of lineups are non-positive for season. The same kind of trends (which I consider shocking, or did til recently) seen for great, good, mid-level and awful teams. There are differences in coaching style but on some criteria these guys look a lot alike. Clearly when you make the right move matters and how right it is, how powerful it is to counter all the non-positive moves. It would be great if nbawowy.com allowed drilling down on lineup performance vs good / great teams, certain situations, etc. (I should re-check how much is possible there or at NBA.com or elsewhere. There is more drillng and summing I could do with Basketball-Reference. NBAvalue.com made that possible in public domain in past down to individual stint with time or scripting.) There is more to be unearthed to explain winning and losing lineups and what separates them.

Which teams just happened to or perhaps made the decision to concentrate lineups well above the norm?

Stan's Pistons lead the way on lowest simple count of lineups, using just 137 so far compared to the league average of 253 or just 54% of the average. Hornets at 166 and Raptors the only other barely below 200. Just 3 teams made it a bit over 35% thru the season with under 200 lineups? That isn't job one but that count is worth knowing and maybe giving reflection, considering how much of an edge teams tend to get from their more concentrated lineups. These 3 teams are not the super elite and have out performed, matched and underperformed expected wins so simple lineup count may not be an immediate pick as a special secret sauce but differences are worth studying further and deeper.

Does Stan's dual titles help streamline the lineup count since he had considerable say in picking the players and how to use them? Probably. But might be worth checking Magic history too. Is this example a potential influence on more teams going to dual title leaders? That is very speculative but it perhaps should be considered.

How does average NBA lineup concentration compare with NCAA, Euroleague, Olympics, other sports? There are other minute based measures of degree of concentration that should be checked beyond simple count of lineups.

As of tonight (per Basketball Reference) 7599 lineups had been used leaguewide. 69% were non-positive, 31% positive. So if your team was at just 66% non-positive, woohoo, your positive share was about 10% higher than average.

872 lineups or about 11.4% of the total have been used the low threshold of 10 plus minutes for season or about 20 seconds per game. 53% of these "bigger" minute lineups were positive.

Separate used over 10 minutes from the under and it is 27% positive for the under 10 minutes group or about half the rate of the over 10 minutes group.

Positive rate for over 20, 50 and 100 minutes doesn't vary much from the over 10 minute group but they are all equal or mildly higher. About 5% of lineups have been used over 20 minutes for season, 1.5% over 50 minutes, 0.6% over 100, 0.3% over 150.

For the group over 150 minutes though 64% are positive. Those used over 5 minute per game for season are on average almost 20% more often positive than those in the 10-100 minute range. But there is less than one of these sized lineups per team. The few and almost 2/3rds proud. Only the Clippers had two of this size and positive. Half the teams had 2 over 100 minutes but in only 6 cases were they both positive. Precious are the decently played and good.

Average net point performance level by minute tier would be useful but will wait for another day.

Lineup under 10 minutes use so far vs. over 10 minutes on average probably wins (revised based on later post about zero minute lineups) 1 in 3-5 times or less. But who knows or acts on that knowledge much? Most coaches probably think they can beat that average. They appear to try a heck of a lot (probably tens of thousands of stints, unknown at this time # of total minutes) but sadly not so well, guessing from the aggregate numbers.

Alnost 18% of all lineups have no elapsed time. They are still official lineups and can have scoring but it is very rare. I assume it results from staggered entry of players during foul shots or timeouts.

If you eliminate zero minute lineups from the under 10 minute subset, 39% have positive results instead of just 27%. So the gap between that group and mid-minute subsets are about 42% smaller than previously reported. But there is still a large performance gap and looking forward those zero minute lineups are completely unproven and probably will behave like the nonzero / under 10 minute lineups or perhaps even worse on average and likely worse than their actual almost always neutral performance.

Celtics have used 125 lineups in just one game. On average they performed at -2.4 per 100 possessions. Much worse than overall (plus 4) or lineups used more often (a bit higher).

Of all net positive lineups, the average use per game was 1.5 minutes. So it goes like this on average: strong performance for 3 possessions. Welp, time for a change. Surely some or many of these could have been extended beneficially.

Rick Carlisle's Mavericks had a high 73% non-positive performance rate for all lineups. The 5 most used were plus 96. All the rest of lineups come in at -115 net. 60% of the 386 lineups saw use in just one game. It would appear that greater concentration of lineup use would help here too.

I'll probably get around to checking Stevens' lineup management vs (Sagarin) top 10 teams compared to overall. His win% against them was less than 10% in yr 1, 21% last season and now at 25%. 18% for career. Seems like it be awhile or take a notable change to get to 40% or 60%.