APBRmetrics

Mike G wrote:

Greg Popovich really costs his team 0.7 points per 100 possessions

Of course he doesn't. You're probably about 2 points off the mark.

Back on track! For the record, I am not saying anything about Greg Popovich. I'm saying that there's some splainin' that needs to take place, first, as to why coaches appear to contribute -1.0 points per 100 possessions, on average, to team outcomes, rather than what I should think is everyone's null of 0.

Whether GP is 0.6 points better than this low average or another figure all together is another question and a good one.

If you manage a company which over time has happened to attract about 97% of the world's most competent professionals in the field;
and if you then rank your employees relative to the average (sales, production, whatever) in your company,
and then you go around telling 63% of them they're creating a negative contribution, because they're below average among the elite,
then 63% of your people will think you are crazy.

If you then tell some more employees that they're barely contributing anything, because in fact they're just a few % above average among the elite in their profession, they too will think you are crazy.

You may then have a small % of people you can cajole that they are worth their keep, because they're noticeably better than nothing, and they too will not be especially appreciative.

Average is not "zero contribution". In the NBA, average is very, very, very, very good.

You can call anything "zero". A one-on-one game between LeBron and Wade? One guy will be a lousy player, if "zero" is immutable. And where do you find a replacement-level DWade?

Mike - I'm not sure what your point is there. Is it that NBA players have fragile egos and constantly need to be reminded that they're amongst the 500 or so best people in the world at what they do? Or is it that they have no concept that when we rate them, we only care about how they stand in the NBA, and so the fact of the matter is that half of them must be below the median? J.J. Hickson is, by all accounts, a fairly terrible NBA player. Would I want him on my college team? Probably. My college IM team? Definitely. Sadly, Hickson plays in the NBA. So he is terrible at his job, which is NBA player.

If Hickson or anyone else is an NBA median player, one of the top 180 or 200 in the league, he is not making a "negative contribution" when he plays. If he were, he wouldn't be getting any minutes.

Some make a greater contribution, but it cannot be true that 60% of players are creating negative point differentials.

Why is that Mike?

Let's say we have a 2 on 2 matchup. Team 1 is Player A and Player B. Team 2 is Player C/D.

Player A is great, +5. Player B is below average, -1. Players C and D are -2. That's 75% of players with negative differential.

Why is what?

If you and I play LeBron and Wade, we'll get beat. Even if we play the best we've ever played.

Is that the same as saying we contributed a negative interaction? Or did those other guys do that to us?

Imagine being the other guys. Did 2 losers just hand it to us? Or did we kick their asses?

Aren't we talking about mathematical ratings? I mean, I understand your philosophical perspective, but mathematically, you can have more than 50% with negative ratings (unless my math is wrong, of course).

I've got another modest proposal, this time for a VORPy system that should satisfy adherents and my principled objections. I call it VORS: Value Over Replacement Schtevie. I will flatter myself greatly and have it be defined as only and exactly 100 points below beautiful zero. 100 points is easy to remember. An easy addend for the extra VORPy math operation. Arbitrary, but no matter, it's easy to remember, which is most important. And every player in the NBA and his momma can feel that he is making a huge positive contribution, none scoring less than a 90. So, everyone gets at least an A!

But never mind this. I want to know why the average coach in the NBA has an RAPM of -1.0. And secondarily why coaches appear to have rather little influence on game outcomes (in terms of the rather limited range of RAPM estimates).

I don't know that a range of 5 or so points per game, from best coach to worst, is very small -- given that they aren't actually on the floor, and that their main function is just to decide who enters and leaves the game at any time.

Athleticism isn't a factor, so there many be 1000 times as many potential coaches from which to choose the top 30 who have head NBA jobs, as there are elite players.
Unlike elite players, when your coach seems to be not so good, there's always a line of replacements at hand.

The lambdas for players and coaches. offense and defense are different. I wonder if there would a negative bias to coaches and a decided edge of coaching impact on the defensive side if the lambdas were all the same. I don't really follow that part of the construction well. I also wonder if finding the "best" lambdas for players was given a higher priority than for coaches.

Popovich was estimated the 9th best coach on a 5 year version of RAPM for players and coaches a bit less than 2 years ago. Maybe a 10 year time period and maybe even using a prior informed basis particularly hurts his rating. He was the coach of the 2003, 2005, 2007 champs but none since. His rating for 2002-2007 might be at the very top.

One stylized fact that leaps from http://stats-for-the-nba.appspot.com/coaches is that one is not an elite coach without having shown a positive influence on the defensive end.

Looking at Tom Thibodeau, whose first place status is based upon one year's work, in terms of the simple before and after, the Bulls' Offensive Rating improved 4.8 and the Defensive Rating improved 5.0. And according to the filter that is RAPM, TT was responsible for -0.8 of the former and 3.2 of the latter.

Why RAPM clearly shows coaches (generally) to subtract offensive value is an interesting mystery.

Crow wrote:The lambdas for players and coaches. offense and defense are different. I wonder if there would a negative bias to coaches and a decided edge of coaching impact on the defensive side if the lambdas were all the same. I don't really follow that part of the construction well. I also wonder if finding the "best" lambdas for players was given a higher priority than for coaches.

My understanding is that the larger the lambda, the more 'squishing' toward the prior there is. So offensively coaches are smushed more than players, but defensively players are actually smushed more than coaches. I don't think it means that coaches are going to be generally moved to the negative as a rule, but I could be wrong.

schtevie wrote:..
Why RAPM clearly shows coaches (generally) to subtract offensive value is an interesting mystery.

A coach can't generally improve his team's offense as much as another good player can. He's only about as impactful as a median player.
Meanwhile, it may be that players don't use their talent to play sound team defense, as instinctively as they work together on offense. Coaching helps a lot on D, and maybe it's largely substitution patterns: yank a guy who isn't working on D, and insert someone who will.

Not so mysterious.

"If Hickson or anyone else is an NBA median player, one of the top 180 or 200 in the league, he is not making a "negative contribution" when he plays. If he were, he wouldn't be getting any minutes."

Why are we assuming that coaches are perfectly rational? The data (especially on Hickson) says quite otherwise. I would love to hear a rational discourse on Hickson typically displaying "non-negative-contribution."

Hickson has been waived, so he's not one of the top 6-7 players per team, aka top 180-200 in the league.
It's not "perfectly rational" to play only your >0 players all the time. The -1 and -2 players (and coaches) are also contributing positively.

If you have a leaking boat, you try to bail water as fast as it comes in. Even if you can't keep up with it, it's better to be bailing than to stop bailing. You have a better chance of staying afloat 'til you reach your destination.

The better your crew, the better your boat will move, against whatever elements. That zero/average line is completely arbitrary.
When an NBA team gets to the playoffs, they meet stiffer and stiffer competition. Their point differential drops and likely becomes negative.

This doesn't mean their contributions had less value; it means the same contributions failed to be better than those of some greater team.