89Hen wrote:
Your definition of proper is what's preventing you from doing it. You didn't like using the GoHens pool because it only had 25 points each week, but that's what we have and those are the teams we're talking about for the playoff spots. Who gives a rats ass if a computer might be able to pick the Valpo/Campbell game while a human might not be able to because they've never seen either play?
I suppose we could try that. But another thing you'd have to do is either pick one person in the pool you consider to be good or do something like present some kind of consensus of the group each time. It has to be a one vs. one comparison with the one in each case being defined ahead of time.
The reason is that if you do something like say you're going to compare one power rating system to whoever does best among a group random chance favors whoever turns out best among the group.
To illustrate: Suppose the system is really better than any particular person in that they would beat any one particular person 60 percent of the time over an infinite number of games. If you compare 30 people to the system over 100 games there is a better than a 99% chance that at least one person will beat the system. There is a 95% chance that at least 7 will.
So you would need to pick your "champion" ahead of time; prior to the comparison. That or maybe do something like pick the team the majority of the people in the poll pick to win in each case.
Why not just compare how the computer does vs the field. IOW, the human participants are ranked each week and through the season based on win/loss and we can just see where the computer falls in the standings.
