[doublec]

Hey, all you people smarter than me in stats, I need your help. I've lurked in the think tank a bit, but just now getting to where I don't feel stupid asking questions.

Anyway, for the sake of curiosity, I plotted team offensive efficiency (O. Eff) and defensive efficiency (D. Eff) against winning percentages in the NBA. Here's what I got.


Offensive Efficiency:

Winning percentage can roughly be modeled as 3.64*(O. Eff)-3.28
R^2 = .62

Defensive Efficiency:

Winning percentage can roughly be modeled as -3.56*(D. Eff)+4.2
R^2 = .41

(Note that my efficiency ratings are for possessions, rather than 100 possessions)

My question is this:

I know that correlation doesn't demonstrate cause, but does the much higher correlation value (R^2) as well as higher coefficient (3.64 vs 3.56) of O. Eff mean that teams with a good O. Eff are more likely to win than teams with a good D. Eff, or is it just kind of a fluke?

[sharpcat]

I would say it depends on what season you pulled your data from, a lot of times trends like this come and go dependent on what types of schemes are being used throughout the league. If you pulled that from this year I would say it holds some weight but I would not count on it every year.

I don't follow NBA too closely but if you look at the NFL these types of trends change every several years, jut look at the past few years with the introduction of the wild cat. A lot of times rule changes between seasons can have an effect on things like this also.

[doublec]

This is from this year.

[sharpcat]

I would say that it would be a useful tool to add to your handicapping this season, just beware this trend could change next year or the year after if they change any penalty rules or anything crazy like that. Also I would only use this as an addition to my handicapping and not just strictly as a way to make picks.

GL

[doublec]

Thanks sharpcat. I'll see if I can find some way to incorporate this into the ole' spreadsheet.

[maxdalury]

why not run a regression with both oeff and deff? You should also get more than one year of data though.

[doublec]

As in (O. Eff-D. Eff)? I have. Obviously it has a very strong correlation with winning percentage, but that's not really as advantageous or unexpected as the large discrepancy here (I think ). And if I understand it, that's essentially the Pythagorean winning % formula.

[maxdalury]

a multiple regression with both oeff and deff as different variables

[doublec]

Ok, I think I got it right here. Here was my end result.

Win %= 3.35*O. Eff - 3.14*D. Eff + 0.28
R^2 = .94

[talnted]

interesting information in here gentlemen, never really thought about this

[uva3021]

Yeah your coefficients and supplementary added numbers won't be something that is constant from year to year

Typically the PYTH formula is used to predict winning percentages over the course of the season and game to game, with an exponent b/w 12 and 17

W% = (Points scored^x) / (Points scored^x + Points against^x)

Its essential that your numbers are formulated into a standard score for this process, so just use the Oeff and Deff and adjust for tempo to get an expected score for each team

For the chances of winning a particular game, you use what is called the log5 formula, another Bill James contrived calculation

W% = ( Team1 % - Team1 % * Team2 % ) / ((Team1 % + Team2 % ) - ( 2 * Team1 % * Team 2 % ))

Which seems reasonable considering Team1% + Team2% will in all likelihood be greater than 1

I believe that's the formula, that's what I had in an old excel sheet from last year when I was messin' around with this stuff

An easier formula is just a linear formula, add .500 to the differential winning percentage between the two teams. This is really only a more desirable method if both teams hover around .500 either way. The further removed the numbers are from .500, the more the linear formula and the log5 formula can demonstrate a considerable margin for error.

Getting me excited lol I love this shit, here is Lakers/Bobcats for tomorrow with homecourt added to Bobcats' chances of winning (4 pts, 11 percent)



I think 43% induces a line around 3 to 3.5, SBR has a nice convenient calculator that should help you find out what the line should be

[doublec]

uva, thanks for the information and the clarification . I'm not really concerned about these numbers working in the future, just this year. I knew the pythag theorem has the exponent, but I didn't know you had to base it off points. I'll definately had to look into your winning % theorem. Looks like very useful stuff.

Is your 4 point home court adjustment universal or team-by-team?

[runnershane14]

how well does that stuff work uva3021? ps..did you use to live on rugby road?

[uva3021]

I just used the 4 pts on average given to home team

[Sinister Cat]

Dave Berri, using data from 91-92 to 06-07, found that it was:

Winning Percentage = 0.480 + 3.138 * Offensive Efficiency – 3.119 *Defensive Efficiency
Adjusted R2 = 0.94

http://www.wagesofwins.com/WinsModel.html

[doublec]

That would be good for general use, but I'm trying for things in this season in particular. Good info though. Uva, don't most teams have a different home court advantage? All my stats are for home/away.

[uva3021]

well you can easily adjust for home court advantage, just use a team1 road % vs team2 home % instead of their season win % in the less accurate yet still reliable linear formula

[doublec]

Hmm. I'll look into maybe combining that somehow. Thanks for all the help guys. Anybody else with interesting NBA formulas, feel free to jump in.

[Peregrine Stoop]

That would be good for general use, but I'm trying for things in this season in particular. Good info though. Uva, don't most teams have a different home court advantage? All my stats are for home/away.

problem is that you will have trouble having big enough samples for unique home courts

in addition, this stuff doesn't stay very consistent year-to-year

You're better off 'winging it' when getting team specific home court

anchor and adjust IMO... anchor on the league average and adjust slightly for:
-how close the game should be
-what kind of environment surrounds the game
and such stuff