Loading ...
Sorry, an error occurred while loading the content.

1274Re: [APBR_analysis] Re: Ballhogs - Dominique Wilkins

Expand Messages
  • bchaikin@aol.com
    Sep 12, 2002
    • 0 Attachment

      The same way Stratomatic, etc. reproduce numbers.  They don't use a
      touch factor.  From economics (and other stat sciences), there are
      always a lot of different ways to _explain_ past results using all
      sorts of models.  Dave Berri's estimates of individual wins reproduce
      team wins very well.  Does that mean his model of individual wins is
      right?  No.  I'm pretty sure that your #'s are ballpark -- they
      mostly reflect subjective ideas that we have.  But I want to know how
      big that ballpark is.

      have you ever used the software? the free download at members.aol.com/bchaikin? its not like i'm hawking the damn thing - its free. but i think you are really missing the boat here because you are basically saying that unless i watch every game and chart every single possession that the way i calculate poss fact and player attributes can't possibly be that accurate. and then you got on to say, in essence, that anyone can reproduce past results, and there are a number of ways to do it accurately. mind you now, you are basically saying all this without even having taken a serious look at the parameters in action, meaning using the software..

      well what the B-BALL software does, and what it was specifically designed to do, is to not only reproduce past results with extreme accuracy, but also to predict future performance based on player's past stats, and this it does. your next question is either - how, i'dlike to see that? or a statement like "...i doubt it very much...". well again, if you haven't use the software, its easy to criticise. you are putting the burden of proof on me, but not looking at the proof available, meaning the software. but i will try to convince you with an example...

      in 91-92 and 92-93 david robinson scored between 23-24 pts/g. then in 93-94 his scoring jumped to 29.8 pts/g. why? what specifically happened? could it have been predicted? can you yourself explain it?...

      here's what happened. a frontcourt player who played a ton of minutes yet had a very low possession factor (touches/min) was traded to the spurs and played in every game but three and 38 min/g - dennis rodman, taking the place of players with much higher poss facts. starting PG avery johnson played in 92-93 (high poss fact of 1.70) but not in 93-94 (don't remember why - injury, traded away?). in 93-94 PG vinny del negro started and had a low 1.25 poss fact (low for a starting PG). starting G/F dale ellis had his poss fact drop from 0.69 to 0.59 in a year playing major minutes. sean elliot (92-93) and willie anderson (93-94) were somewhat similar players and played similar minutes...

      so what happens? david robinson has his poss fact shoot up by 23% - in one season! from 1.14 to 1.40! that 1.40 poss fact was the highest ever touches/min for a starting center in the league until shaq's 2000-01 season. if you look closely at robinson's  stats in the stats database at www.apbr.org, you'll see his player attributes (%shoot, %fouled, %TO, %pass) in 92-93 and 93-94 are virtually identical, meaning what he did once he got the ball was basically the same, i.e. he didn't shoot more often or score from the line more often per offensive ball possession. he just handled the ball 23% more often...

      fine..that's the history lesson. but could this have been predicted? the answer is definitely yes - this is what i did for 5-6 years in consulting. if you use the B-BALL software using the 92-93 players (not the 93-94 players mind you), and trade rodman to the spurs and set him up in the sub pattern to play 36-40 min/g, reduce dale ellis' poss fact by about 10%, and put in del negro at starting PG instead of avery johnson, and run a few similar seasons using the 92-93 data (again s.elliot's stats and w.anderson's stats were very similar), and you'll get similar statistical results to the 93-94 spurs with robinson's poss factor increasing by 20%-25% as well as his scoring going up to 27-28-29 pts/g from 23-24 pts/g...

      you don't even have to take my word for this - you can do it yourself (the free download is the 86-87 to 88-89 teams, you can use that for similar examples or i can send you - for free - the above mentioned seasons). but it works - like a charm....

      any other simulation you ever seen able to do this? i've played them all (or did 6-7 years ago) and none were even close. so while you may believe that many can reproduce past seasons with extreme accuracy, how many can do the above? i mean you are not a typical basketball fan but one who has done some serious statistical analysis. how can you explain the above scenario unless the model works to a "T"? unless the parameters truly parallel the real world? the bottom line is that the possession factors and player attributes are right on the money - if they weren't the simulation would not reproduce real life numbers with the accuracy that it does, nor could it do the above example. its like a jigsaw puzzle, unless every player (the ones with the major minutes) handles the ball as often as he did in real life, and then did with the ball what he did in real life, and shot the same FG%s, how could you possible get the accurate results?...

      other examples could be trading dantley from the jazz to the pistons and seeing his average drop alot without changing his player attributes because he's traded to a team with players with higher possession factors than the team he was traded from, or jeff malone traded from the bullets to the jazz, where his poss fact dropps about 25% -  there are many other examples...

      Can we break it down for the Finals?  Since Stu has directly observed
      numbers from the 2001 Finals, that would give a sense for how
      accurate it really is in a small sample of games.  If you can, do it
      for all players in the Finals so that we can compare with Stu's
      counts.

      if you send me the 76ers stats (or just iverson's) for each of their playoff series that season, i'll calculate his poss fact and player attributes and post them...

      The sorting and limiting work fine.  My resolution is definitely
      high.  I cannot read a lot of the columns, primarily in the basic
      fields (minutes, fga, etc.).  The last number gets cut off
      frequently.  Because I can't cut and paste, it's hard to put in a
      format that I can read the way I want to read it.  Sometimes, you
      want to see %shoot next to minutes and FGA, but you can't do that. 

      yes you can, just double click on all the columns in between the ones you mentioned and the columns are hidden, that way you can see the minutes, FGA, and %shoot columns right next to each other...

      I understand your desire to keep it all proprietary and that limits cut
      and paste and display options, but that also limits usage.

      would you let users get the stats in ascii files for free if you had developed this - typed in the stats by hand over the years?..

      did you know that all - and i mean all - the websites that list complete historical baseball stats (1871-1993 and later) on the internet, including the few that have since gone defunct (like www.baseballstats.com), all got their stats in ascii files from the same single source - not a single one typed all their stats in by hand nor researched the data initially? all the data came from a Total Baseball CD that came out in the early 1990s where they put on it all the stats in ascii files (i found out later accidently/mistakenly). its easy to tell because they all have the same original errors as the Total Baseball CD. so all these people "giving" away free historical baseball stats on the web didn't do the work of typing it all into spreadsheets - so noble of them to give them away for free. moron here has been typing stats into spreadsheets since the days of 8" floppies and computers running CPM, the predecessor to DOS (thinking i'd be using them for creating simulations) never thinking about the internet and these databases...

      i just spent 2 years creating what i believe is the only historical football stats database on the internet that people can using for free. should i give those stats away for free? actually a week does not go by that i don't get emails saying i have no right to "keep" the stats for myself, that i should let people download the entire databases in computer/ascii format for import to spreadsheets because the stats are public domain. but the vast majority of people who use the databases and realize the work behind them are like msg_53@... who send emails simply saying thanks for making them available. and again i'm always looking for ways to improve the databases so suggestions are always welcome...

      bob chaikin
      bchaikin@...


    • Show all 21 messages in this topic