Charting Duke vs. UNC (II)

[Jumbo]

Apologies for the delay, but if ever there was a game that demonstrated the flaws in plus/minus analysis, this was it. The bench guys just didn't play enough to give us a significant sample to judge. And obviously Scheyer and Singler weren't the worst two players on the floor -- they just played virtually the entire game. Anyway, on to the numbers ...

Individuals
Greg Paulus 15-13 (+2)
Brian Zoubek 15-15 (0)
Elliot Williams 65-70 (-5)
Lance Thomas 53-58 (-5)
Gerald Henderson 61-67 (-6)
David McClure 14-20 (-6)
Jon Scheyer 69-79 (-10)
Kyle Singler 63-73 (-10)

Per 40 Minutes
Greg Paulus +13.3
Brian Zoubek 0
Elliot Williams -5.3
Lance Thomas -6.5
Gerald Henderson -6.9
Jon Scheyer -10.3
Kyle Singler -10.5
David McClure -40

Individuals
Paulus-Scheyer-Williams-Singler-Zoubek 5-2 (+3)
Scheyer-Williams-Henderson-McClure-Thomas (x2) 5-3 (+2)
Paulus-Williams-Henderson-Singler-Zoubek 2-0 (+2)
Scheyer-Williams-Henderson-Singler-Zoubek (x2) 5-5 (0)
Scheyer-Williams-McClure-Thomas-Zoubek 3-3 (0)
Paulus-Scheyer-Williams-Singler-Thomas 2-2 (0)
Scheyer-Williams-McClure-Singler-Thomas 0-0 (0)
Paulus-Scheyer-Williams-McClure-Thomas 0-0 (0)
Paulus-Scheyer-Williams-McClure-Singler 0-0 (0)
Paulus-Scheyer-Henderson-McClure-Singler 6-9 (-3)
Scheyer-Williams-McClure-Singler-Zoubek 0-5 (-5)
Scheyer-Williams-Henderson-Singler-Thomas (x8) 43-50 (-7)

[Saratoga2]

Clearly the plus/minus analysis fails to capture the true nature of the game, and because of these results, calls into question the real value of your effort. I appreciate how much you put into the work and it is a great effort, but it appears that some additional ideas are needed to obtain a truer evaluation of player contribution in a game.

Paulus did not play well when in the game, while Scheyer was close to perfect. Somehow, it would seem the evaluation would show that.

[DevilHorse]

I'd like to offer an interpretation.

I think that the way to interpret the first 2 statistics is not what the individual does while they are in the game, but what the team does when the player is in the game. If I recall correctly the numbers for individuals are the team's score minus the other team's score, while the player is in the game. In the case with a small player rotation, we actually have a better read on what the substitute's contribution is than other games, with more substitutes, because there is a more homogenious team that the sub is plugged into.

So, does the sub provide an a net spark or a net lull when they come in? Based on the statistics, it suggests that in this game Paulus gave the team a big lift. Zoubek also gave the team a lift (after all, the team's net score is negative, but with Zoubek in the game it was 0). With McClure in the game, the team (inexplicably) did very poorly.

McClure is a puzzle for this team. Although he is effective on defense, on offense he spends much time on the perimeter where he is virtually ignored and leaves the rest of Duke playing 4 on 5 (or 4 on 4.5). He gets attention only when setting picks and his player might need to be involved in a switch.

Larry
DevilHorse

[CDu]

I'd like to offer an interpretation.

I think that the way to interpret the first 2 statistics is not what the individual does while they are in the game, but what the team does when the player is in the game. If I recall correctly the numbers for individuals are the team's score minus the other team's score, while the player is in the game. In the case with a small player rotation, we actually have a better read on what the substitute's contribution is than other games, with more substitutes, because there is a more homogenious team that the sub is plugged into.

So, does the sub provide an a net spark or a net lull when they come in? Based on the statistics, it suggests that in this game Paulus gave the team a big lift. Zoubek also gave the team a lift (after all, the team's net score is negative, but with Zoubek in the game it was 0). With McClure in the game, the team (inexplicably) did very poorly.

McClure is a puzzle for this team. Although he is effective on defense, on offense he spends much time on the perimeter where he is virtually ignored and leaves the rest of Duke playing 4 on 5 (or 4 on 4.5). He gets attention only when setting picks and his player might need to be involved in a switch.

Larry
DevilHorse

I think you're reading too much into it. I'm suspecting that Paulus has a solid plus/minus because he got more time against the UNC reserves and limited time overall. I doubt it was due to any spark he provided. It's possible, of course, but unlikely.

The more likely scenario is, quite simply, that plus/minus isn't a very useful measure when you're on a losing team. Losing teams have negative differentials, so if you spend more time on the floor for a losing team you're more likely to have a negative differential. The less time you spend on the floor for a losing team, the more likely it is that a couple of baskets for your team can sway your plus/minus to the positive. That's what appears to have happened to Scheyer and Singler (who I thought were clearly our best players) and Paulus and Zoubek (who played sparingly and likely against the UNC subs).

Note as well that, in your comparison of Paulus, Zoubek, and McClure, you're looking at a difference of a couple of baskets. Paulus had a 15-13 score. Zoubek's was 15-15. McClure's was 14-20. That's a difference of a couple of random made three pointers. For any player who plays limited minutes, random baskets (for either team) can greatly skew the results.

Plus/minus definitely is a next step in measuring a player's quality, and I commend Jumbo for his work in providing these numbers. But we should be careful to avoid trying to glean too much meaning from the plus/minus of any one game. It's just too sensitive to random variation, and doesn't account for the quality of the opposition on the floor when a certain player is on the floor.

[Jumbo]

Clearly the plus/minus analysis fails to capture the true nature of the game, and because of these results, calls into question the real value of your effort. I appreciate how much you put into the work and it is a great effort, but it appears that some additional ideas are needed to obtain a truer evaluation of player contribution in a game.

Paulus did not play well when in the game, while Scheyer was close to perfect. Somehow, it would seem the evaluation would show that.

With all due respect, did you read my first paragraph? Or have you been following what I've been saying all year? It's very tough to draw impressions from a single game from a plus/minus perspective. But as the season wears on, the numbers become more meaningful. If they weren't meaningful, teams wouldn't use them to evaluate players and lineup combinations. But thank you for "calling into question the real value of [my] effort." What exactly are you qualifications to do that again?

[Jumbo]

I think you're reading too much into it. I'm suspecting that Paulus has a solid plus/minus because he got more time against the UNC reserves and limited time overall. I doubt it was due to any spark he provided. It's possible, of course, but unlikely.

The more likely scenario is, quite simply, that plus/minus isn't a very useful measure when you're on a losing team. Losing teams have negative differentials, so if you spend more time on the floor for a losing team you're more likely to have a negative differential. The less time you spend on the floor for a losing team, the more likely it is that a couple of baskets for your team can sway your plus/minus to the positive. That's what appears to have happened to Scheyer and Singler (who I thought were clearly our best players) and Paulus and Zoubek (who played sparingly and likely against the UNC subs).

Note as well that, in your comparison of Paulus, Zoubek, and McClure, you're looking at a difference of a couple of baskets. Paulus had a 15-13 score. Zoubek's was 15-15. McClure's was 14-20. That's a difference of a couple of random made three pointers. For any player who plays limited minutes, random baskets (for either team) can greatly skew the results.

Plus/minus definitely is a next step in measuring a player's quality, and I commend Jumbo for his work in providing these numbers. But we should be careful to avoid trying to glean too much meaning from the plus/minus of any one game. It's just too sensitive to random variation, and doesn't account for the quality of the opposition on the floor when a certain player is on the floor.

Exactly. What we can't track is who was also on the floor for UNC (way too much work to cross-reference that). But Paulus, for instance, didn't even play in the second half. Duke was up one in the first half. As you mentioned, he got to play against UNC's reserves. Big difference.

[CDu]

With all due respect, did you read my first paragraph? Or have you been following what I've been saying all year? It's very tough to draw impressions from a single game from a plus/minus perspective. But as the season wears on, the numbers become more meaningful. If they weren't meaningful, teams wouldn't use them to evaluate players and lineup combinations. But thank you for "calling into question the real value of [my] effort." What exactly are you qualifications to do that again?

I agree with most of your paragraph, except for the bolded sentence and the last sentence.

I was under the impression that teams are seeking much more complicated measures (as discussed in the Battier piece) to evaluate players. Plus/minus was, I thought, more of a mainstream media adjustment. And certainly the "sabermatricians" are working to come up with something more complicated and predictive, right?

I agree that Plus/minus can be useful on a larger scale like the course of an entire season. But it has it's limitations: for example, I'm not sure it's entirely useful on losing teams. I could be wrong, but I would have thought that teams would try to look deeper than plus/minus.

[-jk]

Gently, folks. Jumbo does +/- out of the kindness of his heart.

No metric is perfect. The box score provides the most primitive - and often misleading - analysis. Tempo free stats, efficiencies, and +/- are better in many ways. None is perfect, none does well in a vacuum, and none does well in a small n, which a single game will always be.

Let's thank all those who provide these analyses while asking nothing in return.

-jk

[CDu]

Gently, folks. Jumbo does +/- out of the kindness of his heart.

No metric is perfect. The box score provides the most primitive - and often misleading - analysis. Tempo free stats, efficiencies, and +/- are better in many ways. None is perfect, none does well in a vacuum, and none does well in a small n, which a single game will always be.

Let's thank all those who provide these analyses while asking nothing in return.

-jk

I feel like everyone in this thread appreciates Jumbo's efforts on this, and has said so. I certainly appreciate his efforts, and wouldn't want to undertake the workload to do it! My only point (and I don't think Jumbo and I are in disagreement here) is when others try to make too strong an inference on the basis of the single-game plus/minus numbers. When considered over the course of a season, the metric can be a nice tool in conjunction with other stats to measure player's value. But as with any metric, if used incorrectly it can lead people to come to faulty conclusions. So my problem is not with the metric but more with some of the uses of the metric (not by Jumbo, just to be clear).

I hope my last post didn't come across as badmouthing plus/minus. I do think it's a valuable next step, and agree that it can be more useful than simply eyeballing the box score. I am just under the impression that team's are seeking more complicated methods of evaluation than plus/minus at this point.

[Jumbo]

I was under the impression that teams are seeking much more complicated measures (as discussed in the Battier piece) to evaluate players. Plus/minus was, I thought, more of a mainstream media adjustment. And certainly the "sabermatricians" are working to come up with something more complicated and predictive, right?

Of course they are. But simple plus/minus is often the starting point for some of those metrics, and tells at least part of the deeper story they are looking to uncover.

[pfrduke]

Exactly. What we can't track is who was also on the floor for UNC (way too much work to cross-reference that). But Paulus, for instance, didn't even play in the second half. Duke was up one in the first half. As you mentioned, he got to play against UNC's reserves. Big difference.

He came in first at 14:15. Lawson, Ellington, Green, Thompson, and Davis were on the court. At 12:24, Frasor and Hansbrough came in for Green and Thompson. He went out at 9:44.

Second time he came in at 6:02. Lawson, Ellington, Green, Hansbrough, and Zeller were on the court. He wet out at 4:04 and was done for the day.

So never the full starting lineup, but not a lineup heavy on bench (such as Drew, Frasor, Davis, Zeller, and say Ellington).

Not that this necessarily means anything, statistically speaking, just providing facts.