Comments on: Losing Your (Inefficient) Leading Scorer Hurts Your Team

By: Nathan Walker

Nathan Walker — Sat, 28 May 2011 21:36:50 +0000

Usage curves explain that a player's PERSONAL efficiency decreases when usage increases. Whether or not his TEAM's efficiency increases depends on the player, but we can predict it with their usage. Eli Witus found that lineups containing players with high-usage will fare better than lineups containing players with low-usage.

The general trend I have found is that (Dean Oliver) efficiency * usage is the most significant descriptor (rather than separating the two) and produces the best predictions.

By: David

David — Fri, 27 May 2011 21:33:19 +0000

#29 Agree. And I'd love to see this for above and below average players for PPG, REB, and AST. Or maybe by terciles. I think a larger context would help.

By: Bill

Bill — Thu, 26 May 2011 21:42:48 +0000

How James/Wade/Bosh came together is unique, but three star players on one team is certainly not unique in NBA history. Did Larry Bird ever play on a team with fewer than three HOFers?

By: Anon

Anon — Wed, 25 May 2011 17:05:32 +0000

"Not that this will get answered, but I am a WOW type, and I have to admit, I don't get usage curves. The premise is obvious. The more you shoot the less efficient you should be. The less you shoot, the more efficient you should be." This is based on some key assumptions, however (like running/playing an offense that maximizes your play, etc.). Not to mention that the James/Wade/Bosh trio is something that is unique in NBA history.

By: Vjl110

Vjl110 — Wed, 25 May 2011 17:04:13 +0000

#28 - I think using Winshares or anything else to predict the affect of an "average" player would simply replicate Berri's method. The virtue of using a group of randomly selected high-mpg/starting players as a control. You don't need to make any assumptions about what features contribute to wins. This would be a really easy addition the Neil's study, and would give us a really good look at the value of "inefficient" scoring.

By: David

David — Wed, 25 May 2011 15:24:05 +0000

#26 - Neil did a recent post on usage curves. Well, not exactly, but he looked at how per minute stats change when minutes change. This, to me, is the heart of the matter. There really was no conclusive take home message. You can find players who get better or worse and also those that do not meaningful change. I've seen this elsewhere too. I also wonder where this conventional wisdom stuff come from. I know Hollinger likes this idea. But there is simply too much scatter for this idea to, say, be a parameter for player acquisition by the FO. Maybe I've missed _the_ study on this too. #24 - Yes, this is huge. A meaningful null should not be "no change". And I would still love to see the following: Replace every team's leading scorer by an average player using WinShares and simulate the current season. What does eff diff look like then? The conditional analysis Neil did certainly points in the right direction but there is too much variability in terms of how a team makes up for the missing chucker.

By: Owen

Owen — Wed, 25 May 2011 14:55:15 +0000

"wasn't basically right where it should have been at 59.9%." should read "was basically right where it should have been at 55.9%."

By: Owen

Owen — Wed, 25 May 2011 14:53:56 +0000

"I think this shows that WoW simply does not understand usage curves." Not that this will get answered, but I am a WOW type, and I have to admit, I don't get usage curves. The premise is obvious. The more you shoot the less efficient you should be. The less you shoot, the more efficient you should be. But to me it rarely works out as predicted. For instance, shouldn't Bosh have seen a huge bump in efficiency this year playing next to two stars? He was more efficient from the perspective of turnovers but his scoring efficiency actually went down to it's level of two years ago. You look at Kobe's numbers and the pattern is counterintuitive as well. He basically has been in a ts% band of 54.4% to 56.4% for most of his career (all but two years) despite changes in usage, shot attempts, teammates etc. His highest usage year (in his age 27 season granted) wasn't basically right where it should have been at 59.9%. If anything though his efficiency seemed to increase with higher usage (although adjusting for age probably would flatten things out). Clearly there is some relationship between usage and efficiency, but it strikes me as being very weak.

By: yariv

yariv — Wed, 25 May 2011 01:50:12 +0000

As #24 noted, and as discussed in the original post, the problem is comparing losing the leading scorer to losing any high MPG player (I would prefer "high MPG" over starter). Any intention to work on this?

By: Vjl110

Vjl110 — Tue, 24 May 2011 19:08:39 +0000

The fact that losing a starting player (who happens to be an inefficient scorer) lowers efficiency differential 1.2 points really isn’t interesting by itself. Absolutely no negative change is not an acceptable null hypothesis for this analysis. That said, there are ways you could get at answering the proposed question using this data-set.

1)
Try randomly selecting 100 starting players and see how losing those players affects efficiency differential. Now you have a control group which gives you a null value to compare against. If losing a random player is no different in effect than losing an inefficient scorer, that inefficient scoring isn’t necessarily important. If losing inefficient scorers hurts the team more than losing a random starter, you now have much better evidence for the import of inefficient scoring.

2)
Alternatively, build a regression model using some composite metric like WP48/WS/PER for each player, and another using shots/usage or some combination thereof. Then try using these two models to account for the team efficiency differentials that you found. The results of this study wouldn't be as clean as the one above, but they would still be interesting. If usage or shot attempts does not explain the strength of effect on efficiency differential within the above group of inefficient volume scorers, then I am not sure the study supports the value of inefficient scoring.

This is a great study, but I don't feel like I have learned much until you at least find a better null value than "no effect of losing starting player".