Berri 12/28 on Bogut/Bucks

[Joana]

I'm ok with Rosenbaum's work also since three years ago he was incredulous that Redd would get a max deal since Rosenbaum had him as the worst defensive SG in the league by a wide margin and a huge reason why the Bucks defense was terrible. More confirmation bias there.

Hmm... a curious thought: if Rosenbaum's study was to be replicated for this season, Redd would probably be rated as an above average defender. I'm pretty confident that would be the case. I can't avoid wonder what would be your reaction then. You'd stop trusting Dan Rosenbaum's work? Would your opinion on Redd's defensive prowess change? Would you opt for not having an opinion (this could prove to be wise)?

[REDDzone]

"Gaining possessions is highly valued, the use those same possessions is penalized."

Joana, isn't only the UNSUCCESSFUL use of these possessions penalized?

Also, jw, I recall (I think?) hearing you say something along the lines of PER being a better methodology than berri's work. Care to explain why that is?

(Quick sidenote, for awhile I have been wanting to see Joana/Epi debate the merits of objective statistical analysis, they are obviously both good basketball minds with very, very different opinions on the subject. I'm excited. )

[Epicurus]

Not as much as I think he should. The teams, of course, get huge quantities of numbers, much of which are not available to the public. I assume that every coach is quite aware of numerical tendencies of opponents, players and teams as whole, and use such for planning and game management.

Yeah. From my own experience, things like screen/roll defense, rotations, how to guard this or that player, force him left or right, if the opponent team is left handed or right handed, play more or less wing denial, etc, are heavily based on stats.

As far as a summary statistic, like WinsProduced, etc, I think Terry doesn't accept the concept, believing that basketball lacks the discrete events (i.e., much more flow and interrelationships) of baseball and thus summary stats are not as valid and useful in basketball compared to baseball.

He's right. Besides, what can a compilation stat tell him he can't know by checking the classic box-score stats?

Berri's summary statistic is based upon the available boxscore numbers. Its value add is the weights assigned (on the basis of 10 years' worth of boxscore numbers for all games regressed to respective end of season point differentials) and a cost/benefit consideration of the values. With adjustments based upon position played and a team adjustment, the resultant summation value offers a metric which can be used for comparative objectives (who, according to the grounded theory contributes to wins better or worse). To me that is considerably better than just reading across boxscore lines. A fair question is the non inclusion of certain identifiable events in the box scores, such as charge taking. It would be interesting to study the impact of that inclusion in box scores and it consequence to Berri's algorithm.

[Joana]

Not as much as I think he should. The teams, of course, get huge quantities of numbers, much of which are not available to the public. I assume that every coach is quite aware of numerical tendencies of opponents, players and teams as whole, and use such for planning and game management.

Yeah. From my own experience, things like screen/roll defense, rotations, how to guard this or that player, force him left or right, if the opponent team is left handed or right handed, play more or less wing denial, etc, are heavily based on stats.

As far as a summary statistic, like WinsProduced, etc, I think Terry doesn't accept the concept, believing that basketball lacks the discrete events (i.e., much more flow and interrelationships) of baseball and thus summary stats are not as valid and useful in basketball compared to baseball.

He's right. Besides, what can a compilation stat tell him he can't know by checking the classic box-score stats?

Berri's summary statistic is based upon the available boxscore numbers. Its value add is the weights assigned (on the basis of 10 years' worth of boxscore numbers for all games regressed to respective end of season point differentials) and a cost/benefit consideration of the values. With adjustments based upon position played and a team adjustment, the resultant summation value offers a metric which can be used for comparative objectives (who, according to the grounded theory contributes to wins better or worse). To me that is considerably better than just reading across boxscore lines. A fair question is the non inclusion of certain identifiable events in the box scores, such as charge taking. It would be interesting to study the impact of that inclusion in box scores and it consequence to Berri's algorithm.

See, that's one of the issues that makes Berri's work highly suspect. Berri claims that all of WPs are a result of regressions, that it's all empirical and built from boxscore data. But that's completely false. The relative possession value of rebounds, shots and turnovers are determined by theory. The relative weights of, say, made shot to missed shot are not determined by regression, but by theory. It's arbitrary.

I don't see the importance of charges. It's just another turnover.

[Joana]

"Gaining possessions is highly valued, the use those same possessions is penalized."

Joana, isn't only the UNSUCCESSFUL use of these possessions penalized?

Also, jw, I recall (I think?) hearing you say something along the lines of PER being a better methodology than berri's work. Care to explain why that is?

(Quick sidenote, for awhile I have been wanting to see Joana/Epi debate the merits of objective statistical analysis, they are obviously both good basketball minds with very, very different opinions on the subject. I'm excited. )

Meh... possessions have to be used and nobody can use them 100% efficiently. This is where Berri's model is theoretically flawed, he doesn't value possessions symmetrically. I'll try to find some texts about this issue when I have the time.

I don't prefer PER, I just think PER is more harmless because Hollinger is somewhat more humble and doesn't make the extraordinary and grandiloquent claims that are typical of Berri - like proclaiming the irrationality of anyone else.

I think all these theorically-derived formulas are more or less equally flawed and are generally pointless. The point is: there's absolutely no reason to value WPs over, say, NBA Eff. They're two metrics with roughly the same problems and the same merits. The only difference is that Berri wrapped his formula in a victorious and pseudo-scientific rhetoric and used the team adjustment trick, a very clever move that fools less attentive people.

[Epicurus]

[quote="Joana]
See, that's one of the issues that makes Berri's work highly suspect. Berri claims that all of WPs are a result of regressions, that it's all empirical and built from boxscore data. But that's completely false. The relative possession value of rebounds, shots and turnovers are determined by theory. The relative weights of, say, made shot to missed shot are not determined by regression, but by theory. It's arbitrary.

I don't see the importance of charges. It's just another turnover.[/quote][/quote]

No, you are simply wrong. The relative weights of boxscore variables are determined by regression analysis. If a criticism exists, it is that the whole thing is too inductive, not that it deduces from speculative theory, Have you read his book?????? That might make for a good and mutual starting point.

Yes, a charge is another turnover, but the point is who caused the turnover, as that is a observable contribution of a player. A steal is but another turnover, but it's a contribution tof a player to steal the ball.

[Joana]

I read his book.

Send him an e-mail and ask him how is the ratio of the weights for turnovers relative to missed shots determined.

Points, assists, fouls and blocks values come from regression. Not the rest of WPs.

[emunney]

Send him an e-mail and ask him how is the ratio of the weights for turnovers relative to missed shots determined.

Isn't all of the weighting determined by regression analysis? They aren't determined relative to each other.

[Epicurus]

Yep, em, Berri used each boxscore player variable for the multiple regression and thus each boxscore player variable got its own weight.

"The NBA tracks a variety of statistics to measure a player’s performance on the court. NBA fans are familiar with points scored, rebounds, steals, assists, turnovers, blocked shots, etc… The difficulty with these statistics is that some players excel at some aspects of the game, but not at others. And what we want to know when confronted with players with different skills is how each player impacts the final outcome we observe on the court.

The approach we took in The Wages of Wins is simply to utilize regression analysis – a common technique in economics – to determine the relative impact of each statistics on team wins." Berri

No need to email him, just read what he makes available. For what he is attempting, it would make no sense to chose some boxscore variables and not others.

[Joana]

He's lying, or, to be precise, obliterating part of the truth. He obtains the coefficients for those box-score values from regression. The relative possession value of offensive rebounds and defensive rebounds do not come from regression. For example, what's the possession value of ORbs and how was it determined?