At the end of October I did a regression of the scores of the top 25 players against their place of finish and came up with an anomolous positive sign for the effect of fifth place finishes.

I have done the regression again for the top 25 players in the first week of November and the fifth place effect does not appear. The regression I get for November is

Points = 140F1 + 36F2 + 5F3 - 6F4 - 53F5 - 142F6 + 38F7

where F1 is the number of first place finishes, F2 is the number of second place finishes, etc. The standard errors for the seven coefficents are (31), (23), (37), (39), (49), (53), and (43).

The sign of F5 was positive in the October regression. Why is it not positive now?

Here are a number of hypotheses.

1. Ordinary least squares (OSL) is sensitive to outliers. At the end of October we had some very large scores of some of the top 25'ers. In the current competition, MadHatSam has some outlying percentage values, escpecially for seventh place. He commented on my previous thread that he bails out quickly from bad starts to avoid losing points. The current regession has a postive value for seventh place which disappears if MadHatSam is removed from the data set. So one player's scores can affect the results.

2. Last month I overlooked the fact that player's points are rebated to them on the zero min tables when their score is negative. In the current month's standings, the top 25 are likely to include fewer players who had spent some time on the skids before getting airborne (and getting point rebates for fifth, sixth, and seventh places finishes). By the end of the month, the top players with poor luck at the start will have had more time to catch up. Hence, the "rebate effect" might not show up until the end of the month. The rebate effect causes coefficients to become positive or less negative.

I will continue my regression research.