APBRmetrics

[Mark]

The Defensive PER approach you outline has the advantage of being unified and already weighted- if you are otherwise comfortable staying with PER based weights or are willing to tweak everything to your own. It still uses the performance of the lineup counterpart as a proxy for who the original player was defending and could face that skepticism. Still, that could be the way to go.

Though all the different defensive rating formats have value, I would be interested in sticking with your originally stated box-score basis and avoid using on/off. The team defense incorporated for 30% or 50% or whatever of the score seems worthwhile.

I was thinking of using grade average approximations for shot defense along with the known defensive tangibles steals, blocks, rebounds, fouls but I guess you have deal with assists too. But tying assists to a shot defense approximation might not be that big a deal or modification to my initial suggestion.

Grades for shot defense could, alternatively, just be a separate simplified list. Using it as a part of the PER based approach comes down to whether it would make acceptance and use any easier and that will take more feedback. Rounding the corners of that part might bother some but if this is just a player ranking, rounding that part where the data is vague might help with the step up from mostly offense PER to something more complete though imperfect..

[THWilson]

[davis21wylie2121]

[Mike G]

This is tightening up some:

[Mike G]

Some of the leaders coasted to a finish. Dirk didn't finish on top in eWins or in total productivity rate. To squeeze in the necessaries, I'm combining steals/turnovers/blocks/fouls into one category. Here are the (45) players who got at least half of LeBron's eW total.

[deepak_e]

Mike, how does Tracy McGrady have a higher T score than Kobe Bryant? Is that mainly due to Houston's superior defense?

[Mike G]

Kobe and TMac had almost identical rates of steals, blocks, and TO; Kobe's a bit better scorer, while McGrady rebounds a tad more. The diff is that TMac had 86% more assists, standardized.

While Kobe outscored McGrady, per game, 31.6-24.6 (a 28% advantage), he did so in 14% greater MPG. Then, as you suspect, Kobe played in games in which players scored 11% more PPG (101.6/91.2). This makes McGrady's points worth 11% more than Kobe's, doesn't it?

McGrady shot only .507 eff% (to Kobe's .567). This advantage to Kobe boosts his 'effective scoring' by (.567/.507)^.5 = 1.058, relative to TMac. This happens to account almost exactly for the difference in their Sco rates.
_________________
40% of all statistics are wrong.

[Nikos]

Just out of curiosity why do you use exponent .5 in your effective scoring?

Is this a better scaler? Maybe I should have used this in my equation? What is your logic for using it?

[supersub15]

Mike, are you going to do ratings for the playoffs?

[Mike G]

Playoff rankings are coming.

Nikos, the exponent varies a bit from year to year and in playoffs. But 1/2 is pretty normal. It's the number I use when I adjust past seasons; my traditional exponent is 1.

This is one of many variables that I slide around to achieve best fit between player rates -- which translate to eWins -- and team pyth-expected-wins. In big samples like whole seasons, it's always close to .50. In small samples like playoff series (4-7 games between possibly crazily-matched opponents) it can vary quite a bit. In last year's Finals, it approached zero.

No-correlation between individual TS% and winning doesn't mean it isn't better to hit your shots. It just means that in certain environments, the ability to get a shot (and make a fair share) is more vital than not-missing shots. Winning teams have players who can score.
_________________
40% of all statistics are wrong.

[Nikos]

E-wins are very interesting. But are they impossible for you to apply to your historical database prior to 2005? Aren't you missing some stat components that keep you from making a comprehensive historical database where each player has E-wins?

BTW out of curiosity, I have Bernard Kings 1985 season as one of the most productive scoring season in NBA history. The top one I got was Jordan in 1987, but many of MJ's other scoring seasons in his prime are below Bernards of 1985, and even Dantleys in 1984. But most of your MJ scoring rates seem clearly higher than both. How did you derive that?

[Mike G]

I have 15 player-seasons looking better in Sco than Bernard's '85 (including Dantley '86), and another 3 before AD '84.

In '84, Dantley's TS% (.652) and Pts/40 (32.4) were both near their zenith. But he did this in games of 115-114.

BKing in '85 played in 105-110 games. Actually, I'm not sure of this, since he missed 27 G. Maybe his games were even higher scoring, at least for the opposition.

Jordan's got the 4 best scoring seasons, in my list, and 8 better than any by AD or BK. As late as '96, he had a better year than anyone before or since. Since those Bulls gave up a mere 93 PPG, I see that as the league scoring average in games in which MJ participated.

30 points for the '96 Bulls is worth 114/93= 1.23 times as much as 30 points with the '84 Jazz. If we play a game to 114, we will all score more than if we play to 93.
_________________
40% of all statistics are wrong.