APBRmetrics

[schtevie]

Approaching the season's halfway mark, let's check in on the Celtics.

Employing my preferred prism, Adjusted +/-, it appears that the Cs are in a very precarious situation, if the ambition is to repeat as champs. Furthermore, it seems that the implications of this statistical approach have not yet permeated the conventional wisdom.

Doing the math, these are the primary observations:

(1) There is little difference in the aggregate contribution of the starters between this year and last.
(2) However, there are important differences in who is contributing and who is not, within this group.
(3) The bench is worse than last year.
(4) The foremost problem of the C's is point guard play.

Taking these in order.

(1) Last year, by my calculations ,the starting five produced about 11.9 net points per 100 possessions. This year, the number has improved slightly to 12.8 (though this improvement reflects starters playing a larger share of possessions. Holding last year's minutes constant, the productivity this year is actually slightly lower.)

(2) The constant production of the starters as a group, belies big changes in where the positive contributions are coming from. The most important has been the utter collapse of Rajon Rondo and the resurrection of Jesus. Both are surprising. Rondo, supposedly on the up-side of his career, has become the third worst Adj. +/- guy of all starters in the league. (And note the likely understatement in that Aaron and Steve estimate includes some of last year's - somewhat positive - contribution.) And to think he was being touted as an all-star! Weighting by his minutes played, he contribution has gone from +1.4 last year to -7 this year, a huge negative swing. Making this up however has been Resurgent Ray, going from -0.9 to 8.0.

A point of digression. The fact that an aging Ray Allen can reclaim past glory so spectacularly, suggests to me that nagging injuries are the most important factor in determining the variability in player productivities. Apparently, last year, he was simply hurt throughout, never having fully recovered from off-season ankle surgery (I think?). And more relevantly for the Cs' immediate prospects, the accumulated nicks and accompanying relative decline of Paul Pierce (last year's 6.7 to this years 3.4) do not bode well for the post-season.

As for the other two starters, KG is actually way ahead of last season, contributing 10.5 vs. 5.9 last year (though recall that last year averaged similarly phenomenal numbers early on with many games of both playing injured and then playing back into shape) and Perk is roughly the same, -2.0 this year vs. -1.1 last.

(3) If the starters as a group are similarly productive, it is the bench that must be worse (-4.3 this year vs. -0.6 last year). This is not however because of the loss of Posey, whose contribution last year was a team low -3.3. Leading the negative contributors after Rondo is Eddie House, dragging down team productivity by 4.1 points per 100 possessions. Which brings me to the next point.

(4) Between Rondo and House, the vast majority of which are point guard minutes, their effect on team productivity is to decrease it by 11.2 points per 100 possessions (whereas last year they were a combined net positive of 2.7 points). Solve this problem and the prospects of the Cs are vastly improved. And though this puts the Marbury issue in stark focus, the more pressing question is: has Sam Cassell's game deteriorated so much that he could not be providing some relief until an alternative solution is reached? Even last year his Adj +/- numbers were low positive. And goodness knows that he is rested.

And it seems to me that there are other reallocations of minutes that would serve the Cs well. As he has always been, Tony Allen is an extremely good player. Why he has been so under appreciated all these years is a bit of a mystery, but surely one aspect has been that he was always in the shadow of Paul Pierce. Well, now it seems to me that Paul Pierce needs a rest to (hope to) get back to form and Tony Allen is playing better than him (says Adj. +/-) anyway. Let Pierce spend some days on injured reserve, so as to recuperate for the playoffs, and let's see what Allen can do.

Similarly, Leon Powe should soak up some of Perkins' minutes whenever the relative size issue would not be catastrophic. He is playing extraordinarily well this year, and this could really help the cause.

I don't know if there are any free agents out there (Starbury aside) that are likely to help the Celtics' cause this year. To repeat, it seems mandatory that at least a band-aid (tourniquet?) be found for the point guard problem, then there needs to be some really good bench coaching, in terms of allocating minutes to productive players, and finally, and perhaps most importantly, there needs to be a whole lot of luck with injuries. Hopefully Pierce will rebound (health-wise) and none of the other aging stars will go down with injuries. If so, there is a chance. Otherwise, it will be one and done.

[bastillon]

To me it's completely different. What the Celtics need most is shooting/spacing. Their bench has one 3-point shooter(House). No suprise why their production went down so much. With this kind of spacing no team can succeed in today's league. This case is tough from statistical point of view, because there's really no way yet, how to determine good spacing by numbers.

Tony Allen and Rondo are maybe the worst shooting duo in the NBA right now. And THIS is why they have to add another shooter. Without any doubts Marbury could help.

Rondo is one of the most overrated players in the game. Nobody is mentioning those double-teams he lets opposing teams to send on Big Three. Rondo is highly-ineffective off-the-ball, because of his inability to shoot, so the Celts are trying to get the ball in his hands, so his defender wouldn't disturb their offense as much on help defense. I'm advising you to look, how his defender behaves in situations, where Rondo doesn't have the ball in his hands. In the offseason I thought he would improve jumper much and he did the opposite. There's always E.House to replace him, but his ballhandling inability and not-too-great defense makes Rondo even better. Sam Cassell doesn't look like he's gonna play this year. Tony Allen can't play the point. That's why Marbury factor would be so important. Not to mention he's still able to play bball and of course his will must be much improved given his CY.

I also don't agree with you on the Posey thing. By losing James they lost another key shooter, who really made everyone better by improving spacing. +/- stuff won't tell you, I think, that he was Pierce's backup mostly, so it would be tough for him to do well in this rating. I wonder how his +/- could be that low for another reason. He's not hurting the team in any way, his basketball IQ is as high as you can get, he brings spacing and yet he's great team defender. It's common-known that +/- has some liabilities. I think it's the one. By the way is there any new, this-season-included adj. +/- ? Would be interesting to see how Posey's numbers are looking now, in NO.

[Mountain]

I don't think I saved the 2007-8 season adjusted ratings by themselves. With basketballvalue's move to 2 yr basis which is useful, the 1 year values for that season or this one have been removed- wish they were still there for reference.

Not sure if the values you cite are from basketballvalue or a run yourself but accepting them I'd still say of course that we don't know for sure if the values from single seasons for particular players are close to true impact.

Perhaps Ray Allen has surged that positive this season (putting aside the issue of a small contribution from last season) or maybe the surge isn't quite as strong and rating is higher than true. Same in reverse for Pierce this season maybe he is too low. I also think he is saving it for the playoffs to some degree. Garnett last season may have scored lower than true and Rondo higher than true. Their quality of play may have changed too but we can't be sure how big the errors are on particular players in a season and in the resulting season comparison.

If you go by the 5 season figures the Celtics have been / will be? carried by Pierce and Garnett and Allen has not been a strong adjusted +/- force. But he may be different right now, given the context and his current performance. We don't know. Multi-year had Rondo as -5. I previously mentioned he had one of the biggest differences between raw +/- and adjusted this season. He is in a very very favorable context. But adjusted still labels him a significant drag. The all-star hype is mostly missing this level of analysis.

I'd still prefer a statistical/adjusted blend of some kind, either after the adjusted is produced or use the boxscore better at the input stage to give more credit / blame where more credit / blame is due. Equal credit/blame to all input assumption on each play is still losing significant available accurate in my mind detail in the name of picking up the uncounted effects. I think a method that tries to balance the counting of the seen and the unseen by some method is going to be better. Being completely agnostic to the seen at play level except as it influences the season level findings is going too far in my mind. I'd rather apply some judgment and tweak it over time with input from many than stay agnostic to the seen.

Dan Rosenbaum's move to "Overall +/-" with statistical at about a 80% weight and pure adjusted at about 20% makes a pretty strong case to me that pure adjusted isn't sufficient on its own. Adding back 80% of the boxscore makes a pretty strong case to me that equal credit/blame isn't the right basis for credit/blame.

[Mike G]

[schtevie]

The issue of the imprecision of point estimates of Adj. +/- aside, I don't understand why there is recurrent confusion about the concept and implications of Adj. +/- itself.

It seems to me that the basic answers to Mike G's questions are straightforward.

(1) Would teams have more success by just not playing anyone in the above list? If they had a better alternative (that is another player with a higher Adj. +/-) the answer must, on average, be yes (at least in a marginal analysis). Yes, there might be particular match-ups or game situations where one could extract whatever is (relatively) positive within an over-all negative average rating, but this doesn't overturn the general rule that below-average players are below-average players.

That said, the NBA is not the Lake Wobegon Basketball Association. Not everyone can be above average. Restrict rosters to only those players who currently have positive ratings, and some fraction of these, not too different from half, would become the negative contributors.

(2) Why do teams whose starting PGs appears as negative also have a backup who's negative? It is a fact that the "average" PG in the league is a negative contributor. Eli W's penultimate post on countthebasket.com show a positional average of -1.2 for point guards, the lowest of any position (SG = -0.3, SF = 1.0, PF = 0.3, C = 0.5). Also, check out the table of Steve and Aaron's most recent article at 82 games and one sees only 21 PGs with Adj. +/- ratings over zero. There simply aren't enough "good" point guards to go around. But why is this so?

It is a plausible and non-controversial claim that, in general, there are positive returns to height as well as to athleticism in basketball. A less elegant title for the 'Point Guard' category is: 'Relatively Short But Quick Guy Who Is Less Likely To Have the Ball Stolen From Him When Dribbling Up Court Than A Taller Guy And Who Also Is Relatively Adept At Passing The Ball To Teammates In Advantageous Positions'. That the league on average may be misappraising the relative value of quickness vs. height in stocking this player category seems possible, and this might be what drags down the PG average relative to other positions.

Finally, in a similar vein, let me reply to Bastillon's defense of James Posey. Again, conceptually, Adj.+/- is the only measure which in fact includes spacing, shooting, team defense, Basketball IQ, and every other factor one can imagine. One may not like the result of any particular regression. One may argue that the imprecision of the estimate obscures the true value of the player (for example, his rating has improved in New Orleans). But one cannot argue that it does not include the very factors that it does include.

[Mike G]

Perhaps we shouldn't assume anyone is more 'confused' here, but try to answer some questions, while acknowledging what we do know. PG is perhaps the one position that is virtually always present on the court. You may have 2-3 on the court, for short spells; but almost never do you have no PG out there.

A team may run without a C for much or most of a game. A team can be competitive with 3-4 F's or with 3-4 G's.

In football, some great players -- receivers, running backs -- are only productive due to the presence of a quarterback. In basketball, the PG is the 'passer' who allows everyone else to be 'positive'.

A below-avg QB or PG is still entirely necessary to the productivity of the others of the team -- offensively. Of course, the PG is required to play defense. So this must be where the 'negativity' of his Adj+/- comes from.

The (minority of) 'positive' PG's must be giving up less on defense than they provide on offense. But to whom are they giving it up? To players other than PGs, apparently -- otherwise, the PG population wouldn't rate a net negative.

It must be that bad teams tend to use more PG's (at one time) than good teams use. Or that below-avg PGs are more often used in tandem.
_________________
`
39% of all statistics are wrong

[Mike G]

[DLew]

Given that the general convention is to define the zero level as average, then yes, by definition half of the player-minutes will constitute below average, i.e. negative performance.

If one decided to make zero equal to replacement level then fewer than half would be negative, and if you made zero equal to the worst player in the league then no one would be negative.

The scale of adjusted plus-minus is entirely relative and so the idea of negative performance should not concern you.

[Neil Paine]

It's a zero-sum game, Mike. Seems like we've had this chat before...

[basketballvalue]

[Mike G]

Thanks, Aaron. That goes a good way to explaining this anomaly. Have the Lakes also used Fisher/Farmar together and done badly?

[schtevie]

What conceivable advantage is there in trying to define the performance of replacement level players as the zero mark? If you were the mother of a below-average player, such a norm might make you feel a little better - assuming you didn't understand the definition - but seriously? Zero has a particular meaning in a zero-sum game. Rate higher and you are improving your team's performance on the court. Rate lower and you ain't. (Usual disclaimers about all else equal.)

I am all for letting a thousand flowers bloom, but assigning the zero mark to a notional replacement level player (which is arbitrarily defined anyway) is simply a bad idea. It does not promote clear thinking; it obscures it. This little world of ours needs clear, intuitive, and ultimately useful standards and terminology. What is the counterargument?

Anyway, back to the Celtics etc. Could someone suggest a list of the free agents available for easy pick-up by the trading deadline (i.e. not those requiring multiple players/salary cap contortions to effect their transfer).

[royce.toyfu]

Replacement level players are available, average level players usually aren't. If you are thinking how someone is relative to average you may know they aren't in the top half the league, but that doesn't mean you can find someone better. If you can accurately approximate replacement level then if you have someone below it there should be options available to actually replace them.

This goes right to the Celtics question you ask. If the players you want to replace are below average but above replacement then it won't be easy to improve by replacing them.

There was one row I saw on that spreadsheet marked FALSE for F or F that had Farmar as player 1.

Other than that it looks like Bryant must be handling the ball, is there any way to tell who these lineups were used against? Was it in response to a Tall Ball lineup by an opponent where Bryant was able to take advantage of a slow point?

[Kevin Pelton]

Bryant had to serve as backup PG this week because both Farmar and Sasha Vujacic (who has been filling in for Farmar) were hurt.

[bastillon]