1. #1
    Waterstpub87
    Slan go foill
    Waterstpub87's Avatar Become A Pro!
    Join Date: 09-09-09
    Posts: 4,043
    Betpoints: 7224

    Historical Injuries

    I am attempting to backtest a model, but with the NBA individual players are very important. Is there any website that would have players status listed for particular days?

  2. #2
    brewers7
    brewers7's Avatar Become A Pro!
    Join Date: 03-11-06
    Posts: 298
    Betpoints: 4441

    Why do you need their status when you can see whether they played or not?...

    Players' statuses on game days can be very inaccurate anyway...

  3. #3
    ljump12
    ljump12's Avatar Become A Pro!
    Join Date: 12-08-09
    Posts: 108
    Betpoints: 258

    Quote Originally Posted by brewers7 View Post
    Why do you need their status when you can see whether they played or not?...

    Players' statuses on game days can be very inaccurate anyway...
    +1 Use the boxscores.

  4. #4
    Waterstpub87
    Slan go foill
    Waterstpub87's Avatar Become A Pro!
    Join Date: 09-09-09
    Posts: 4,043
    Betpoints: 7224

    Quote Originally Posted by brewers7 View Post
    Why do you need their status when you can see whether they played or not?...

    Players' statuses on game days can be very inaccurate anyway...
    Because that would implied that I have access to future information when I am testing out of sample.

    Getting a Questionable, Doubtful or probable allows me to weight a formula.

  5. #5
    brewers7
    brewers7's Avatar Become A Pro!
    Join Date: 03-11-06
    Posts: 298
    Betpoints: 4441

    I see your point...

    Problem for you is this:

    I have been following NBA injury reports daily for at least 15 years and "Questionable", "Doubtful" and "Probable", don't pan out at all..."Probable" is accurate at least 90% of the time, but I have seen "Doubtful" players play more than one might think..."Questionable" is basically ridiculous as that tells me nothing and if the player is important enough to make a difference in the line, then I stay away from those games if a major/key/line-mover player is "Questionable"...

    Doubt if you will find historical information on this... For my database, I have whether the player played or not, which is good enough for me as you could run your "future information" 2 ways:

    1) test it without the player in question playing
    2) test it with the player in question playing



    Quote Originally Posted by Waterstpub87 View Post
    Because that would implied that I have access to future information when I am testing out of sample.

    Getting a Questionable, Doubtful or probable allows me to weight a formula.

  6. #6
    DAC
    DAC's Avatar Become A Pro!
    Join Date: 10-15-15
    Posts: 1
    Betpoints: 12

    Funny you should mention that, I joined this forum to ask the exact same question.

    I have spent countless months developing a model getting it to the point of adding a significant advantage to the public odds/lines only to realize the NBA doesn't actually have to announce who is playing, most of the times you only find out after the game has started. As an Australian I cant tell you how bewildering this is, it is beyond ridiculous.

    Anyway, as you mentioned, creating a model at the team level as opposed to the player level is useless. The problem with box scores is, you are using information (who ended up playing) that isn't known until the end of the game. So as it stands my model is useless, because i have no way of knowing what the lineups listed in the box score are until after the event!

    The only way you can accurately replicate in and out of sample is to have each teams list with an injury status for each player so your lineup projections in sample and out of sample are the same.

Top