Statistical solution to autobenching
I was annoyed during the Maryland game when Zavier Simpson sat on the bench for what seemed like almost 10 minutes after he picked up his 3rd foul, and our lead turned into a deficit. As has been mentioned on this site many times, it seems that Belien (and all coaches) conflate the goal of avoiding time on the bench with the goal of avoiding fouling out. He likely sacrificed many minutes of playing time for Zavier as a result, and made a game that otherwise likely would have been a comfortable double digit win into a game that was much closer. Interestingly, Belien proved this strategy wrong, IIRC, when he allowed Isiah Livers to keep playing after he picked up an early foul against Nebraska when he was essentially forced to due to Charles Matthews' injury. All Livers did is finish with a double double and that one foul, playing 34 minutes.
So let's stop complaining and do something about it. I propose that one of you with the skills and a lot more time than me take this approach:
Set up a linear regression estimating the number of minutes per foul for each player in a particular game based on:
- the M player
- the opposing team
- the referee team
- home or away
You could even develop and validate the model in separate datasets to show Belien it works.
But then you could take it a step further using a Bayesian approach with this regression serving as your prior probablity, and update the posterior probablity in real-time during games informed by the number of minutes played and fouls for each player. Belien could then see that when Zavier gets his 3rd foul, or Isiah picks up his 1st, they likely have X number of minutes left of playing time, and Belien can decide whether to spend those minutes earlier or later in the game.
The auto bench is annoying and sometimes deployed in excess IMO. One variable to consider here as well is that Coach B uses the auotbench as a punishment. He obviously is foul averse (and for good reason - puts your players at risk and present opposing team with free points essentially) but it seems at least in part that this long time for X on the bench may have been a reminder that his foul was dumb. (Which it was).
You can argue the merits of that as well - who knows - it may be the threat of the autobench is what prevents additional fouls and thus makes the guys less likely to foul out over the remaining minutes.
Would love to see the model above though!
Good effort, but statistical analysis alone is unlikely to work here, or at least what you described here. Unless Belien is willing to participate in this analysis by altering his benching policy, no amount of historical data will help you. There is severe endogeneity or self-selection problem here. Players know Belien’ policy and it is affecting their behavior. On the other hand, picking early fouls affects how a player plays after that. Does he become tentative? Does the opposing team try to go after him so that he picks up another foul? And so on.
However, if Beilein decides to participate and were to announce that for the next 10 games he will have a different benching policy, then you might have enough data to do such an analysis.
"Basketball is not a sport. It is a statistical network."
I would enjoy reading such analysis, but believe that there are too many variables. What coach/team/players is going volunteer to be the control group?
Seems to me this Belien guy does not know how to coach.
Boy no fucking kidding. It's amazing how the board picks up on random, meaningless things that obviously bother Brian (like Beilein's policy on auto-benching) and then it becomes a "cause celebre" amoungst the plebeians for while.
We have the best coach in college basketball on our sidelines. I'm pretty sure he knows more about basketball than Brian, any of us and 99.9% of his peers. If he thinks the auto-bench policy is correct I'm going with him.
I am a basketball coach - middle school and high school level at a fairly competitive level for the middle school and meh division for high school. I am not a professional coach - I do it purely as a volunteer, but I have done it for almost a decade and I have compiled a very good record. I know about 1/100000000 of what Beilein knows, but possibly more than Brian in terms of in-game playing coaching, and here are my thoughts on the auto-bench.
Generally speaking, with a few caveats listed below, it works. There are two main benefits: (1) It ensures that if the game is competitive in the second half, and more in the 4th quarter, you have your best players available; and (2) it is actually a deterrent against fouling. On the second point, since I have shifted to something of an auto-bench philosophy, my players know that they need to play disciplined and smart, otherwise they will end up on the bench. They avoid going for "glory" plays on defense (flying in from 10 feet away to go for a block that 9/10 times results in a foul, reaching in on drives that usually result in and-one situation, etc.
Here are the few situations in which I deviate from the auto-bench:
1. If the player being benched is my best player (most of my teams tend to have one or two stars who are a cut about the rest), and I see the game getting out of control with him on the bench, I will deviate to stop the bleeding, but I always tell him that if he picks up another, he has to come out for a long while. My view is that it doesn't do much good to have all of your guys available in the second if the game is essentially out of contention by then.
2. If I have 2-3 players at the same position who are relatively equal. On my middle school team, I have 3 kids who play the 4 spot who were almost all mirror images of each other. Depending on game flow, who was playing well that day, etc., I relaxed the auto-bench for that particular group, since the drop off from one to the other in the event of fouling out was negligible. If player 1 had 3 fouls at the half, but he was playing a great game offensively and cleaning up the boards, for example, I would sometimes leave him in longer than the auto-bench rules said to.
3. Beilein seems to invoke the auto-bench even after 1 foul depending on the time of the game (a foul in the first minute for example). I generally don't apply it until 2 fouls.
It's worked really well for me. I know that the auto-bench is the bball team equivalent of "why no screens, Borges" but given Beilein's record, especially of pulling out wins late in games, I cannot see why Brian or anyone else would even question this.
"Autobench" as a deterrent to committing fouls is a principle I'd never thought of. Maybe Beilein should switch to the "Death penaltybench"; commit a foul and sit out the rest of the game.
I propose we trust the guy with 40 years of experience, who is one of the premier coaches in the sport, and who is the winningest coach in M basketball history, and spells his last name BEILEIN.
I’ll get right on it!
Zavier picked up his 3rd foul, and Livers had 1. If we can keep it close and get to the second half with our best defender on the bench to able to play during crunch time, I'm good with it. Especially knowing that once he got his 3rd, the other team would probably make a concerted effort to attack Z.
Right - Beilein wants to ensure (or close to it) that his leader is in the last 10 minutes of the game. And you know what?; it worked out. He said that if the whole thing went south, he'd have put Z back in earlier. And yes, there is no auto-bench for one foul; yes, we've seen some of that with Teske, but it situational.
Maybe Simpson was so effective late because he had no fear of fouling. Maybe he would have played timid and our defense would have suffered if he stayed in the game. Maybe we won BECAUSE of the autobench.
He had 3 fouls in 21 minutes. He was on pace for a foul out after committing a really dumb foul 40 feet from the basket. No amount of historical data can prove that the next possession Cowan wouldn't have barreled into him and drawn a block call -- Then Simpson sits for about 13 minutes with 4 fouls instead of sitting 7. How would that have worked out?
We don't know what would have happened, but Beilein erred on the side of caution with our most important player. It resulted in a comfortable win on the road against a ranked team. What the hell are we trying to fix?
Thank you for making an important point that has been largely overlooked.
This is exactly correct. I swear, it seems many people here don't understand basketball very well.
Coach Beilein's "autobench" and what happened with Z are 2 completely different things.
One can reasonably argue that benching a guy immediately after picking up a first or second foul early in a game is an overreaction (personally I understand it, especially on a team with a very short bench - a starter fouling out is catastrophic).
After all this time, it's clear JB strongly prefers to keep his players out of foul trouble and playing fully aggressive down the stretch.
It's possible autobenching has cost us a game or two over the years; I have no doubt there are games Michigan wins because opposing players foul out or play tentatively while our guys are unencumbered by foul trouble.
After the Maryland game Coach B was effusive in his praise for Simpson, but specifically said he would talk to him about the dumb 3rd foul. It's a lesson he wants him to learn well, so it won't happen in the tournament.
If Turgeon is any kind of coach, he would absolutely attack Beilein's essential, vulnerable player. Cowan would go right at him on both ends of the court. Simpson likely would either back off and give Cowan space to shoot or drive, and on offense would give up the ball early rather than risk a charge.
I loved when the guy checking me was in foul trouble; it's a huge advantage.
It's easy to say just play your game, and if you foul out so be it; that doesn't take into account your opponent and their adjustments based on your foul situation.
Looking at a player's average fouls per minute is meaningless. If you had a large enough sample size to look at average fouls per minute while in foul trouble, you might have something.
I see a big difference between the autobench when someone picks up their first foul in the first few minutes of the game versus someone who gets their third at the start of the second half. The opponents’ approach to offense seems much more likely to change (attack that player) with 3 fouls than 1. Though also agree with others here that part of this is likely about teaching and changing future behavior; and in general it is probably a good approach to maintain options at the higher-leverage part of the game (the end) so long as you can keep it close up until then (as the announcers pointed out while Simpson was on the bench).
I wish Beilein showed a little more flexibility in his auto-benching, but there is so much wrong about the assumptions here. And BTW, I am sure Beilein believes strongly that his auto-benching of X was reaffirmed by how the game played out.
I'd go and watch Beilein's eloquent explanation for his action on the postgame video; that would be part of doing your homework.
We've gone over a lot of this before, and your instructions to someone else to perform the work (wtf) misses out on a pile of variables, some of which Beilein mentions. The homework part.
Among those you want to add are how Xavier Simpson coming back after a leisurely 11 minutes (at the 12 minute mark is Beilein's standard) and blowing the a** off of the opponent to win going away laughing also figures.
I started out, like you, thinking that all minutes held the same value--a point scored in the 12th minute has the same value as one in the 19th. But what they absolutely do possess are different MEANINGS for the decision-makers, and a decision taken in the 19th is freighted with all the events that came before it. I have yet to see a model that takes this into account, and don't believe it's likely that one can.
You also want to account for the two players who subbed in for X, because a lot of the pants-sh*tting and knicker twisting that the hysterical side of the mgolemmings were so spectacularly performing involved their mistrust of said players. As Beilein said in his postgame presser, he had plenty of confidence they would manage. He enumerates the tasks that they needed to perform to keep the team afloat. They performed them; in fact, Brooks's D was excellent, as people over at UMHoops have detailed in the last few days.
"I started out, like you, thinking that all minutes held the same value. They do, arguably, but they don't. But what they absolutely do possess are different MEANINGS. The meanings we ascribe to them translate into action; that action has value. I'm not seeing anything like an adequate model in your instructions or in any of the other far better analyses I have yet examined. "
This, and the rest of the post, is flowery worded bullshit. You don't actually make a meaningful factual statement in the entirety of this post.
I worked to make that sentence clearer. The question of whether a decision or point made in one minute is equal to one made/taken in another is at the heart of the argument, as you will see if you dig into arguments people have about this, here and at umhoops.
We can only see up to the level of our own awareness; that's a given. Your post just below makes that clear. A complicated issue that over-simple assertions do not settle.
The meaningful factual statement is that leverage matters. Minutes earlier in the game are less important than minutes later in the game, because you know more about the game state later in the game. While it may seem that two points at the end of the game are equivalent to two points at the beginning, it's not true, because your opponent has had the entire game to react to that opening basket -- every decision both teams have made has come in the context of those points having been scored (or not scored).
Yes. There are some interesting counter-arguments, because you shape those later minutes with earlier decisions, including a decision to yoink the guy earlier rather than later. But in the end I tend to side with your argument. It's actually quite a fascinating, multi-faceted problem.
"Minutes earlier in the game are less important than minutes later in the game, because you know more about the game state later in the game."
This is obviously false. A game tied at halftime is more important than when Team A is down 5 with 5 seconds left.
Every minute is worth the same in the long wrong.
The average leverage at the end of the game is higher than the average leverage in the middle of the game. You can cherry-pick examples, as you've done, but your overall point is wrong. It's counterintuitive, but it's extremely clear. If your goal is to win the game, minutes later in the contest are simply more important. It's why one bucket doesn't make much of a dent in the win probability chart in the first couple of minutes of the game, but it can swing it wildly late in a close game.
Every single decision in basketball is contextual, to one degree or another. If Team A loses by two, you can't simply say "if Team A had scored three more points in the first half, they would have won." Clearly, Team B's strategy was dictated by the score, and they would have made different decisions down the stretch in order to try to win.
No statistical analysis is needed. If a player fouls out, they foul out and lose whatever minutes left. If you bench them, the loss of minutes is guaranteed. If you don't bench them, the loss of minutes might happen.
So, to be clear, are you for the auto bench or against it?
against, obviously.
The only reason I asked, (I'm not trying to be a sarcastic prick here) is because you made a couple of obvious statements which were true but did not shed much light on which way you felt.
I personally do not like it really early in the game like Matthews got against State, but I loved what Beilein did with Simpson at Maryland. I thought it was absolutely necessary and obviously it is easy to back up my thoughts because it worked out. Had Michigan lost, I would have still agreed with the decision made, for what it's worth.
I think the one thing people forget about the 'Autobench' is the thought process of the opponent.
I have no clue what the right answer is, and trying to come up with a blanket approach is not really possible either. That said, I get where coaches are coming from because they cannot control what the opponent is going to do.
Let's just assume Z stayed in with his 3 fouls. What is going to prevent Maryland from isolating him one on one for the next 5 possessions (because let's be real, they were not doing much on offense the whole game anyways) and just dropping their shoulder into him on the way to the basket? Any charges, missed shots or other turnovers (in a 5 possession scenario) are going to be incredibly offset by a single blocking foul called on Z. This also assumes Z has the mindset to play defense like he has 0 fouls on him as well. As a coach, would you rather have possibly 4 fouls on Z with roughly 15 minutes left up by 4-5 or have Z with 3 fouls with about 8-11 minutes left down by the same amount?
Each amateur coach can decide that for themselves, but I think the majority of coaches, given our roster this year, would pick the latter simply because you can be assured you can get a rested, max effort (meaning he doesn't have to worry about a foul) Z for 50% of the second half. In the other, you're going to have to white knuckle the rest of the game.
I am a huge fan of stats, Sabermetrics, Moneyball and all that good stuff but with 'Autobench' there is frankly too much to quantify that is not based in stats, such as:
- How is the opponent going to react to an elite player in foul trouble?
- How is the player that is in foul trouble going to react? Is he going to play scared on the defensive end? Is he going to be able to mentally put it aside that he is in foul trouble?
- And the real wild card: How are the refs going to react? Are they more apt to give a player in this scenario the benefit of the doubt or are they going to keep a closer eye on them?
- Does home and away matter?
- Does the talent level of the opponent matter?
Excellent questions.
"Let's just assume Z stayed in with his 3 fouls. What is going to prevent Maryland from isolating him one on one for the next 5 possessions (because let's be real, they were not doing much on offense the whole game anyways) and just dropping their shoulder into him on the way to the basket? Any charges, missed shots or other turnovers (in a 5 possession scenario) are going to be incredibly offset by a single blocking foul called on Z. This also assumes Z has the mindset to play defense like he has 0 fouls on him as well. As a coach, would you rather have possibly 4 fouls on Z with roughly 15 minutes left up by 4-5 or have Z with 3 fouls with about 8-11 minutes left down by the same amount?"
This is more or less exactly what Beilein said in response to the question in the post-game presser. It's available at umhoops.
I believe that a serious, well designed statistical analysis would show that this matter, finally, is 6 of one and half dozen of another. Here, one can legitimately criticize or defend either side 'til the cows come home.
In principle, I oppose "auto-benching". My experience is that attempts to attack persons to draw additional fouls is usually counter-productive...you don't draw the fouls and it disrupts your normal, more productive flow of your offense. Otherwise, it is simply a difference of opinion-pretty nonconsequential.
I strongly disagree that attacking a foul-plagued opponent is counter-productive.
Rarely did I have a good game when in foul trouble. I was never able to ignore the fouls, and trying to play aggressive, sound ball while avoiding fouls is really challenging.
OTOH, the minute the guy checking me was in foul trouble, I smelled blood. I'd go right at him and often had a free path to the hoop. Even better if he tried to stop me; it's a huge win to foul someone out.
IMO, it's a basic, sound strategy to attack your opponent's vulnerability.
But you don't need to draw another foul for the strategy to be successful. You just need an easier scoring opportunity because your opponent is more hesitant. There was a Kenpom analysis on this a couple years ago that illustrated this.
True. The purpose of attacking a guy in foul trouble isn't just to foul him out, it's also to take advantage of a 'weakened' defender.
Pavlov once said "control your conditions and you will see order" unfortunately a basketball game does not end itself well to controlling variables. I do think a within-subject design would be the way to go here though rather than between-group if you could control all the variables.
I love this idea, but I also have a small criticism (that may be targeted at the wrong crowd here... we shall see). Here goes...
Another criticism of Coach B as of late is that he hasn't given his junior players enough live game time to acclimate to the combination of college-level speed and the complexity of his offense. While I agree that autobench can be frustrating, especially when it sees one of the team's best catalysts sidelined, it does allow for some of those junior players to step in and gain some much needed minutes. We are going to need them in the tournament and barring an unexpected slip up from Purdue, we aren't going to win the league anyway.
I posit that we look at autobench as a mechanism Coach B can use to provide game experience for younger players as the team prepares for the B1G tourney and the NCAA tourney respectively. Of course, this shouldn't be the only means by which DeJulius, Castelton and other get minutes, but I think you get the idea.
Thoughts?
Beilein has spoken on this both before and during this season.
In the past, one season, I thought he went too far in the direction of developing the young guys. In my opinion, in 2013, we lost regular season games because Spike got too much playing time and leads were lost or deficits were built. But then, when we needed him, in the first half of the national championship game, he was prepared and ready to go. In Beilein we trust...
So JB isn't afraid of developing younger guys. What's happening now? Specifically this season, he has stated that he's not going to put guys into games when they're not prepared. I read this to mean that the younger players simply weren't demonstrating in practice that they had the full grasp of what the team was trying to do.
The late emergence and additional PT for both DDJ and Castleton in my assessment, is an indicator that those two players have made the progress necessary to earn the coaches' trust.
The auto-bench has been a constant throughout the season (and the past). The difference earlier this year has been who came in as replacements.
Yeah, Coach B has said that hours and hours watching guys in practice and on film impact his decision-making more than a couple minutes of game action.
Certainly some guys perform their best in game conditions, but all the little things like floor positioning, footwork, defensive awareness, etc. show up in practice. JB watches these guys every day, and he's pretty damned good at determining when somebody's ready.
Player development doesn't just magically happen; it's a good bet Colin Castleton is far better now than he was in November.
Why does everything have to be reduced to some form of analytics? Is it just a way to give our subjective feelings (mild annoyance in this case) some sort of objective value?
The whole reason for this call to the calculator is OP's mild discomfort at seeing a professional basketball coach make a decision that he didn't like. I mean, cumong man. We didn't even lose the game.
Take this energy and put it into developing coping strategies to help you deal with things that mildly annoy you--in this case, things you can't control. Once that's done, we can return to enjoying the game and trusting the coach to, you know... coach. He's done a pretty fine job already this season, and there's a lot of hope going into the playoffs. Enjoy the ride.
While folks are at it, could you also do a 50-state analysis of telehealth laws cross-referenced against licensure and supervision requirements for nurse practitioners?
By Friday would be great.
Thanks.
My spouse is a Certified Pediatric Nurse Practitioner-so I concur with your suggested project. Here, at least there would be some objective data to work with.
Fire him.
1) Beilein knows his team more than you, I would hope. He seems to have a knack for this coaching thing...
2) Z needs to rest sometime anyways. Who is to say that our win wasn't a result of Z being fresh for that last run?
3) We won. Great team effort, nice work by the bench in the absence of Matthews.
Yes...that 2 points by the bench was most scintillating!!! Otherwise, as has been noted, the variables of this game renders any scientific analysis null and void...and this is why I love this game!
Well... kenpom.com actually has both "fouls committed per 40 minutes" and "free throw rate" (while not a perfect proxy for all fouls drawn because not all are shooting fouls, it is pretty informative) for literally every player in college basketball.
Assessing who is likely to foul and draw fouls in a particular game is, like you know, what the coaches currently do.
The model and ex-post testing you suggest could be done, but to what end? The tricky bit is getting Beilein to pay attention to your idea.
My leanings tend toward a much softer touch on the autobench policy, especially since we have so many players with such low foul rates (all 5 starters plus Livers are at 3.2 fouls committed per 40 minutes or lower). At those rates, every starter could play 40 minutes of every game and never foul out.
So what gives? Well, a single game is a tiny sample and volatility is the measure of actual results to vary relative to an expected norm. Variations happen. JB has his philosophy and adheres to it. And I'm a believer in BOTH of the following things: 1) I support JB and think he is a great coach, and 2) I think his take on autobenching is too mechanical and could benefit from analysis and adjustment.
That being said, I don't think he's seeking out my opinion on the matter.
There you go...I can concur with every angle on this topic! Personally, as a player and coach, I REALLY liked playing well and winning. And though I played and coached very aggressively...I worried like hell-and this would, at times, yield some "conservative" decisions being made-ones that could be defined as being unfair to the players. I think even an outstanding, confident coach like Beilein succumbs to these pressures.
As far as I'm concerned, a player is maximally utilized if he fouls out with 1 second left. Why that player would spend valuable, useful minutes on the bench to prevent fouling out is beyond me.
Not sure you thought this through. Yes, nice if that last foul is used--say to prevent an easy bucket--at the one second mark. How do you insure that? How do you insure it doesn't happen to X at the 10 minute mark, leaving you without a point guard down the stretch? Keeping him in, as you suggest, hardly insures that.
Everyone agrees that they want X in there, including Beilein. It's maximizing his play that's the key.
--
The hidden bonus here is that a refreshed X played like gangbusters down the stretch. And maybe had figured out several of the actions MD was using while sitting out.
--I have no objection to the fan who questions the strategy; that's fun. But the person who then goes on to assert that it's all down to one or two variables, and that they know and Beilein doesn't. . . ? That's a little insulting, not just to Beilein but everyone's intelligence. The guy has been evaluating and RE-evaluating this question for decades.
The OP has gone off the rails. The game of hoops, as the game of life, is impossible to distill down to graphs and charts. All the regressions or probabilities don't mean squat to organic creatures in the heat of a game.
In Beilein I trust. I assume the auto bench is supported by a preplanned substitution strategy. Who is on the floor and score of game are major considerations. I strongly support the punishment for dumb foul idea.