Wow. This is Cool. Coaches ranked statistically.

[jzmagik]

is the stat nonsense because you aren't bright enough to understand it? Or is it nonsense because you don't like the result?

Was I supposed to understand it after your explanation?

[SolidSnake008]

stats isn't everything...gotta think about things like intangibles

[ElectricMayhem]

I KNEW Michael Curry was a better coach than Pat Riley. Now we've finally got the proof.

[jzmagik]

I KNEW Michael Curry was a better coach than Pat Riley. Now we've finally got the proof.

[Lakeshow2417]

no statistics besides rings and finals apperance can pinpoint who the better coach is.

hell not even wins can define that. din nelson is #1 and I would trust him over flip barely

[Davidm89]

no statistics besides rings and finals apperance can pinpoint who the better coach is.

hell not even wins can define that. din nelson is #1 and I would trust him over flip barely

I don't think rings or finals appearance have any bearing on how good or bad a coach really is. Take Phil jackson, put him on the Nets and they still aren't a contending team.

My personal opinion of basing a coaches worth is their ability to turn a team around if given the ability, making adjustments when needed and the ability to mesh your players together.

[fudgie]

@ the OP, Seriously, you're getting mad about a list that no one knows the measurements for (please tell me I'm just reading into your post--I can hardly tell when people are being sarcastic or serious without joking). If that's the case, find the coach you think is the best and make a stat for all the coaches that will clearly give your favorite coach the advantage. Case close. I'm sure no one cares or even understand this list.

I wasn't being sarcastic.

I agree with this result.

I really believe Stan Van Gundy is the best coach in the NBA.

Do you also believe Jim O'Brien is the second best coach in the NBA?

[KnickelandDime]

is the stat nonsense because you aren't bright enough to understand it? Or is it nonsense because you don't like the result?

Or is the stat nonsense because people who don't know how to use stats in general will miscontrue it to meet their ends?

[Imadogg]

is the stat nonsense because you aren't bright enough to understand it? Or is it nonsense because you don't like the result?

http://en.wikipedia.org/wiki/Face_validity

Any stat should have this at the very least. A stat that puts (Please Use More Appropriate Word) coaches above great coaches just sucks.

[floppymoose]

I understand what plus minus is, and have at least some intuition about adjusted plus minus (thought I've not seen a clear explanation about how that is calculated), but I have no idea how it applies to coaches. Coaches are not players. What does it mean for a coach to have a particular "Offense per 100" value?

[mysticbb]

I understand what plus minus is, and have at least some intuition about adjusted plus minus (thought I've not seen a clear explanation about how that is calculated), but I have no idea how it applies to coaches. Coaches are not players. What does it mean for a coach to have a particular "Offense per 100" value?

As jinxed explained it takes the coach as the virtual 6th man on the court. Every lineup now consists of the 5 players on the court + the coach. I think it has noise, but some of those results are VERY reasonable. Having Scott Skiles as the best defensive coach is something we can expect. Look at the Bulls and the Bucks during his tenure.

Another thing which is somehow in agreement with other observations: Coaches can make a bigger impact on defense than on offense. We know that the numbers are showing that an indivdual offensive player can impact the offense more than an individual defensive player. We now the see the opposite with the coaches. That is also in agreement with common sense, that the defense is greatly impacted by defensive schemes, while the offense can be greatly influenced by an individual player.

Overall the numbers might be surprising, but they are not completly unreasonable. Also, when we look at the results, we should also keep in mind that there is still an error, the error is also different for different coaches. That means: some coaches can be "overrated" by this stats, others are "underrated". Well, usually a 5yr APM model result has +/-1.5 as an error (not quite sure how much the error is here).

[floppymoose]

I still don't get it. Is this a different meaning of plus minus than what I am used to? The plus minus I know requires on court and OFF court time to compare. In the method you describe the coach is always on. What does plus minus even mean in that context?

[bert stein]

remember: the main flaw in using adjusted plus-minus as a stat for players is that it does not take into account complementarities amongst players.

but - one of the key roles for coaches is _exploiting_ complementarities between players on his team (or interaction effects between players on his team and on the opposing team).

think about the implications for using adjusted plus-minus as a measure of coaching performance. this is stretching advanced stats a bit too far.

[mysticbb]

I still don't get it. Is this a different meaning of plus minus than what I am used to? The plus minus I know requires on court and OFF court time to compare. In the method you describe the coach is always on. What does plus minus even mean in that context?

+/- always means the scoring difference. If a team scores 70 points with the player and the other teams scores 60 while the player is on the court, his +/- would be 10 (70-60=10). What you mean is On/Off Court +/-, the difference of the team performance with the player on the court in comparison to the performance without him.

APM is working differently. In that case the authors of those results used a brute force method to solve a set of equations which looked like this:

sum(xo_i*A_i) = points scored

where xo_i is the offensive APM of player i

sum(xd_i*A_i) = points against

where xd_i is the defensive APM of player i

As you might see the more time a player is on the court, the more information I get about that player to solve the equation more accurately. More playing time reduces the error. For coaches being "on the court" always it just means the error is pretty small, because the sample size is big (in comparison to players).

The values are just saying that Coach X is that value different from an average replacement coach per 100 possessions (they write 200 possessions, which just means 100 on offense and 100 on defense which is equivalent to a pace of 100).

[floppymoose]

Ok, thanks for the explanation. I think I get it.

The results make a lot of sense to me. I actually think O'Brien is a very good coach, and his detractors are confusing the lack of talent on his squads with poor coaching.

It certainly seems to pass the sniff test with Skiles, JVG, Fratello, and Sloan all excelling at defense and JOB, D'antoni, Smart, and Nelson juicing the offense. I'm not surprised to see Larry Brown killing the offense as he plays way too structured with not enough freedom for his players to get the opportunistic easy baskets.

Biggest outlier to me are the Rick Adleman values. My eyes tell me he juices the offense - the numbers say the opposite. Maybe it was all those years of Yao. I always thought Yao forced Adleman to get away from some of the things he liked to do offensively.

[mysticbb]

Well, I had the formulas incomplete (no idea why), but whatever.

You also have to take the average values of the coaches into account here. An average coach has around -0.32 on offense and +0.43 on defense (simple average here, not weighted). Overall the average is +0.1. That means in average coaches have a slightly more positive impact than the players in average. Their biggest contribution comes on the defensive end. While a coach with better than -0.3 on offense is a better offensive coach.
Well, that might be just because the coaches are basically not that much different after all in terms of their impact from each other. The differences between the best and the worst in that stats is 5.6 while for the players it is 13.4.

[DSMok1]

Just adding my 2 cents here:

This system is calculating 5+ years average Regressed (via Tikhonov regularization) Adjusted Plus/Minus. So the players are treated as a static quantity, as are the coaches. This will cause some issues because the player's aging complicates matters, particularly when evaluating coaching changes. Rip Hamilton is not the same as he was 5 years ago.

The on/off for coaches has far fewer samples than for players. It basically depends on coaches or players changing teams, whereas for players we're dealing with on/off court, quite a few samples per year.

Thus, one must expect more error for the coach valuations than for the average valuation of the player over the 5+ years. That said, a coach is probably more static (no "aging") in true talent over the 5 years, so the results are more useful in that regard than are the player's "average" RAPM's.

[Chicago76]

If players and coaching ability is considered "static" over a 5 year period, that's more than just a little problem. The maturity/aging curve will have a huge impact on year-to-year +/- for the players. Coaches are considered to be always on the court, so there are no on-off observations for them w/ the exception of coaching changes.

The only way to validate the findings is for the original researcher to go back and run 2,3, and 4 year tests on the 5 year results. My gut tells me there will be huge swings in the results.