Forums

[Hank Chinaski]

Originally posted by pretermitted_child
It looks like your algorithm does not distinguish the following case:

Actual Score: NE: 32; CAR: 29

Entry#1: NE: 30; CAR: 24
Entry#2: NE: 34; CAR: 24

The calculation for Entry #1, according to your algorithm would be: (32-30)^2 + (29-24)^2 = 2^2 + 5^5 = 29

And the calculation for Entry #2 would be: (32-34)^2 + (29-24)^2 = (-2)^2 + 5^5 = 29

This is why, as I had mentioned in my post on the superbowl contest thread, that you need an additional metric for accuracy -- I chose the absolute value of the difference between your guess for the point difference and the actual point difference. In this case, the point difference in the actual game is 3 (=32-29) (which, in some ways, indicates that the teams were evenly matched; a higher point different would, presumably, indicate the opposite). The point difference for Entry #1 would be 6 (=30-24), while that of Entry #2 would be 10 (=34-24). Thus, under my algorithm, Entry #1 would win, which, I think, is a fairer result.

As for the sum of squares method in general, it penalizes guessing in a non-linear manner. This is why I suggested using absolute values.

you don't need absolute values. If you go over you lose, or are you suggesting thats not fair?