2017-18 lineup analysis

Home for all your discussion of basketball statistical analysis.
Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Thu Aug 02, 2018 4:37 pm

The sample sizes are usually small because of the coaching behavior.

The raw or adjusted lineup performance is not a sure guide but imo it is the best starting point for a guess when everyone is guessing. Add subjective on top but start with the performance.

It can be shown that dink lineups on average underperform the most used lineups. Very few teams I have ever checked had dink lineup performance better than non-dink lineups and usually both were good or bad and thus dink lineup performance was not at much advantage. Comprehensive historic studies show the same thing on average.

Respect different competencies and roles but do not overdo it. Imo coaches have been given far too much freedom on lineups and many have not proven they deserve it. I would argue many have proven they DON'T. Look at coaches with majority or super majority of top 5-10 lineups negative (in regular season or playoffs), high dink lineup %s and win% for all lineups or dink lineups in low 30%s or in some cases high 20s. You can have bad luck or bad stretches on a few but there are coaches who have bad lineups all season or for multiple seasons without any change of behavior or result.

I left room for discretionary situational lineups. One might differ on how much room to leave or how strict to be or be totally hands off. I think being hands-off is the real extreme, problematic position that too many front offices take or approximate more than the reverse.

Are you fine with the freedom that resulted in all the lineup result examples in this thread, in / after say January, and bigger minute lineup posts I've made in prior years?

Setting targets with mutual approval is a deliberate, cooperative approach. Coaches making it up on their own, usually in the moment is less than that. It is often not refined, consistent with history or analytics. If some want to continue that way go ahead. I don't. I've seen even the greatest coaches do highly questionable lineup management for seasons and playoff series that probably should have discussed more or flat changed.

My view is that lineup management is not 5-10% "off" (though in a league where 1-2 pts matters and is worth a lot, even this would be significant) but rather it is way to way, way more off depending on team and that the status quo behavior is totally outdated and unacceptable. But that is just my view based on my analysis & experience. fwiw to anyone else.

If someone with 20 plus years of inside NBA experience looked at the data the way I have they should be able to interpret it even better and override it with a higher success rate but I think they'd do better using this approaching along whatever else they do than to not do rigorous quantitative lineup analysis.

Even teams getting acceptable or good lineup results may have sub-lineup inefficiencies that they are not aware of or dealing with sufficiently.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Thu Aug 02, 2018 5:19 pm

First lineup detail mentioned in this thread: the Ferrell, Matthews, Barnes trio for Mavs. Identified as bad in late October. Rick Carlisle goes ahead with it, makes it his 2nd most used trio. -5.6 per 100 possessions. Twice as bad as team's overall bad performance. Pairs within the trio started bad and stayed bad. But hey coach's freedom of choice was great for being bad.

Coaching discretion probably could be higher at start of year. Some super early trends may turn around, especially if something changes. Injuries heal, different play calls / shot distributions, defensive schemes, rookie awareness & execution, etc. Small sample lineup analysis at least causes it to asked if the leaders think it is as good or bad as it has been so far. But as season progresses, there is a stronger basis to pay attention to the results, especially if a team runs a rational lineup use / testing program.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 04, 2018 3:42 am

Blazers 5 most used lineups, avg. performance of plus 4.3 pts per 100 possessions. Everything else (dink and a few slightly bigger) , plus 1.8.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 04, 2018 3:50 am

Raptors most used 5, plus 10.5. Rest of lineups, 5.6.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 04, 2018 3:58 am

Celtics most used 5, plus 8.6 pts / 100 poss. Rest of lineups, plus 2.2.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 04, 2018 4:05 am

There are limits how much players will play overall, so you can't play totally stacked with best player lineups all the time and you have to fill out a full rotation responsibly; but most teams have too low concentration of use of better tested good lineups. Lower than feasible. Often absurdly lower.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Thu Aug 09, 2018 5:34 pm

It appears than only 4 of the top 15 lineups on JE's 5 man RAPM played over 400 minutes last season. Seems like many probable significant losses of potential edge production if they had played more. These were estimated at adjusted plus 9-10 pts / 100 possession. Lineups not used even 5 minutes per game. I'd consider these major coaching mistakes on the surface unless offset by other evidence. A couple teams had 2 or more and that is more understandable / accepted. But generally the coaches knew this information and ignored / dismissed / forgot it or they didn't know it. Unacceptable imo any of these ways. Injuries somewhat impaired a couple but all of them could have played more. And probably should have. And management & analytics didn't get the coach to do that. Too much deference, too much willingness to allow random, inefficient lineups.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 25, 2018 3:30 am

TWolves' starting lineup and 2 one change variations played played over 40% of total minutes are around a plus 8 per 100 possessions clip. The average of the rest of the lineup was around a weak -4. Thibs did poorly with the custom matchup game. Stay basic. Even more. There are minute limitations on guys that constrain use of the top lineups but they possibly could go at least a bit further with them. If not in regular reason, then playoffs. They certain need to run the rest of rotation differently and closer to neutral or better to go anywhere.

DSMok1
Posts: 850
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: 2017-18 lineup analysis

Post by DSMok1 » Sat Aug 25, 2018 10:55 am

Has anyone done an analysis showing that lineups that play together more have benefit "beyond the sum of the players" vs. lineups that play little?
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 25, 2018 11:41 am

That would be worth doing anew and well. Maybe for biggest and biggest five lineups vs rest or next 20 then rest. Using RPM, BPM, PIPM. Whole league, individual teams.

I forget exactly how much was done / said about improved lineup performance with more time. I recall that it was said that it did. Maybe I'll dig in the archive and try to find that study that Cherokee or someone of that era did. First search attempts failed to complete on my phone (and once again is almost certain my phones fault).

Really to do this the plus minus results need to be adjusted for opponents (and maybe other factors). Don't recall if they were in the study in my memory. As usual to do ALL that should be done takes more time.

... sum of individual boxscore metrics would be an inaccurate comparison when the production comes from big and small minute lineups and the whole question is whether minutes of lineup usage matters. Might need separate average big and small minute lineup usage boxscore metrics and / or RAPM to sum up and compare to actual plus minus.

DSMok1
Posts: 850
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: 2017-18 lineup analysis

Post by DSMok1 » Sat Aug 25, 2018 11:55 am

There could easily be selection bias issues--i.e. lineups that do unexpectedly well should, in theory, be given more minutes by the coach. Of course, your thesis is that coaches don't typically recognize which lineups are unexpectedly good or bad.
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 25, 2018 12:13 pm

That "thesis" has been checked pretty often and appears likely to be pretty supported even with raw plus minus data but if someone wants to do that level of analysis or the more advanced with "professional" rigor and discipline, please do. Admittedly my series of posts on the topic are not the end of what could / should be done. I've been trying to raise interest in topic. I don't claim to have resolved it fully in the time I've give it. I have claimed the patterns that I see as far as I have taken it to date. Which is a lot further than anyone else I am aware of in recent times and in public for certain subtopics / approaches. I have opinions that I have shared but I start with being descriptive.

I would be very interested in what you could do on the topic. And others. A big part of my intention behind these postings was to try to get others working on the topics.

Selection bias... how much time do they need to get over other lineups for this to be an issue or major issue? Within the vast world of dink lineups (under 80 minutes for season) I wouldn't think it was a major issue but somebody tell me if it is. Under 10? Under a minute a game or 8 seconds per game seem real small, I.e. not that much "selection" compared to even smaller. The few lineups over 5 minutes per game? Maybe they do need extra study, adjustment or separate treatment.

I've tended to do lineup analysis at season level but more could be done at game and stint level. Compared to other topics I feel that not enough have shown serious interest or done enough on topic. I will acknowledge 5 man RAPM, lineup finders and the stat detail they provide are major contributions, some blogs giving lineups a look and renewed interest in player interactions.

I have doubts about the depth and quality of most teams' work on topic based on what I can see. I'd love to hear from ex-team analysts but such calls almost always generate no response or no revealing response.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sat Aug 25, 2018 1:15 pm

Although I find the number of lineups tried for small minutes extremely high, it is the total time devoted to this practice that matters more. The immediate results do not seem to justify it. The degree of credible learning from it does not seem high or even medium or maybe even low. More concentrated lineups seem to give better results on average and more credible learning.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Sun Aug 26, 2018 1:10 am

I working on something in this vein but it might be awhile.

Crow
Posts: 5567
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2017-18 lineup analysis

Post by Crow » Wed Oct 03, 2018 1:18 am

Where did Kings end up on lineup management?

10 of most used 12 negative, 14 of top 20 negative.

No lineup used for average of 3 minutes per game for season. Only 6 over 1 minute.

643 dink lineups. Only 30% of lineups were cumulatively positive for the season.

Resulting stayed bad from mid-season, even worse in some ways.

Was there a plan? Was there meaningful evaluation / adjustment? I'd have to look further but on top view it looks like a chaotic unsuccessful mess.

Post Reply