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

Expand Messages
  • igor eduardo küpfer
    I assembled the results of all NBA games between the 1992 and 2003 seasons, for each game noting the number of days rest each team had heading into the game,
    Message 1 of 1 , Aug 7, 2004
    • 0 Attachment
      I assembled the results of all NBA games between the 1992 and 2003 seasons,
      for each game noting the number of days rest each team had heading into the
      game, and each team's winning percentage heading into the game (with a small
      adjustment, noted below). I regressed these variables against the
      Pythagorean score of each game and found to my surprise that the days rest
      variable was not statistically significant. This struck my as odd, and
      caused me to wonder if I have done something wrong -- I am not really used
      to multiple regressing, and perhaps I am not doing it right, or am
      misinterpreting the results.

      I present the results below in hopes that someone can point out my error.


      Variables:

      PythScore - Winning team's Pts^14 /
      (Winning team's Pts^14 + Losing team's Pts^14)
      WHome - Winning team Home = 1 or away =0
      Win%_W - Winning team win% leading into game
      Win%_L - Losing team win% leading into game
      Rest_W - Winning team's rest days leading into game
      Rest_L - Losing team's rest days leading into game
      Rest_W2 - Rest_W squared
      Rest_L2 - Rest_L squared
      RestAd - Winning team's rest advantage, ie Winning team's rest
      days minus Losing team's rest days
      RestAd2 - RestAd squared, but keeping the same sign (ie if RestAd
      was negative, RestAd2 would also be negative)

      Note:
      For the Win%_W and Win%_L variables, I added 1.5 games to the win and loss
      columns. That is, at the first game of every season, every team has a
      1.5-1.5 record going into their first game.

      Here are the regression results for the home/away, winning percentages, and
      days rest variables:

      Predictor Coef SE Coef T P
      Constant 0.747300 0.006284 118.92 0.000
      WHome 0.041848 0.002387 17.53 0.000
      Win%_W 0.129275 0.007726 16.73 0.000
      Win%_L -0.127516 0.007206 -17.70 0.000
      Rest_W 0.001036 0.001235 0.84 0.402
      Rest_L -0.002776 0.001253 -2.21 0.027

      I don't know if I'm interpreting this correctly -- the losing team days rest
      is significant, but the winning team's is not? That would be very strange.

      Here is the results, raising each team's days rest to the second power:

      Predictor Coef SE Coef T P
      Constant 0.744968 0.005616 132.64 0.000
      WHome 0.042559 0.002361 18.03 0.000
      Win%_W 0.128975 0.007726 16.69 0.000
      Win%_L -0.127359 0.007206 -17.67 0.000
      Rest_W2 0.0000105 0.0002082 0.05 0.960
      Rest_L2 -0.0002954 0.0002229 -1.33 0.185

      Neither of the day's rest variables are significant, contradicting earlier
      results, I think.

      How about if we just look at the wining team's advantage in days rest, ie
      how many more days rest the winning team had over its opponent:

      Predictor Coef SE Coef T P
      Constant 0.743772 0.005544 134.17 0.000
      WHome 0.042971 0.002321 18.51 0.000
      Win%_W 0.128795 0.007725 16.67 0.000
      Win%_L -0.127331 0.007206 -17.67 0.000
      RestAd 0.0000791 0.0001969 0.40 0.688

      And the same thing, but squaring the rest advantage term:

      Predictor Coef SE Coef T P
      Constant 0.743741 0.005473 135.88 0.000
      WHome 0.043011 0.002322 18.53 0.000
      Win%_W 0.128690 0.007725 16.66 0.000
      Win%_L -0.127316 0.007205 -17.67 0.000
      RestAd2 0.00000580 0.00000558 1.04 0.298

      Again, the days rest variable is not significant.

      --

      ed
    Your message has been successfully submitted and would be delivered to recipients shortly.